AI Research

Simulation-based pipeline tailors training data for dexterous robots | MIT News

Published

2 months ago

July 11, 2025

When ChatGPT or Gemini give what seems to be an expert response to your burning questions, you may not realize how much information it relies on to give that reply. Like other popular generative artificial intelligence (AI) models, these chatbots rely on backbone systems called foundation models that train on billions, or even trillions, of data points.

In a similar vein, engineers are hoping to build foundation models that train a range of robots on new skills like picking up, moving, and putting down objects in places like homes and factories. The problem is that it’s difficult to collect and transfer instructional data across robotic systems. You could teach your system by teleoperating the hardware step-by-step using technology like virtual reality (VR), but that can be time-consuming. Training on videos from the internet is less instructive, since the clips don’t provide a step-by-step, specialized task walk-through for particular robots.

A simulation-driven approach called “PhysicsGen” from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) and the Robotics and AI Institute customizes robot training data to help robots find the most efficient movements for a task. The system can multiply a few dozen VR demonstrations into nearly 3,000 simulations per machine. These high-quality instructions are then mapped to the precise configurations of mechanical companions like robotic arms and hands.

PhysicsGen creates data that generalize to specific robots and condition via a three-step process. First, a VR headset tracks how humans manipulate objects like blocks using their hands. These interactions are mapped in a 3D physics simulator at the same time, visualizing the key points of our hands as small spheres that mirror our gestures. For example, if you flipped a toy over, you’d see 3D shapes representing different parts of your hands rotating a virtual version of that object.

The pipeline then remaps these points to a 3D model of the setup of a specific machine (like a robotic arm), moving them to the precise “joints” where a system twists and turns. Finally, PhysicsGen uses trajectory optimization — essentially simulating the most efficient motions to complete a task — so the robot knows the best ways to do things like repositioning a box.

Each simulation is a detailed training data point that walks a robot through potential ways to handle objects. When implemented into a policy (or the action plan that the robot follows), the machine has a variety of ways to approach a task, and can try out different motions if one doesn’t work.

“We’re creating robot-specific data without needing humans to re-record specialized demonstrations for each machine,” says Lujie Yang, an MIT PhD student in electrical engineering and computer science and CSAIL affiliate who is the lead author of a new paper introducing the project. “We’re scaling up the data in an autonomous and efficient way, making task instructions useful to a wider range of machines.”

Generating so many instructional trajectories for robots could eventually help engineers build a massive dataset to guide machines like robotic arms and dexterous hands. For example, the pipeline might help two robotic arms collaborate on picking up warehouse items and placing them in the right boxes for deliveries. The system may also guide two robots to work together in a household on tasks like putting away cups.

PhysicsGen’s potential also extends to converting data designed for older robots or different environments into useful instructions for new machines. “Despite being collected for a specific type of robot, we can revive these prior datasets to make them more generally useful,” adds Yang.

Addition by multiplication

PhysicsGen turned just 24 human demonstrations into thousands of simulated ones, helping both digital and real-world robots reorient objects.

Yang and her colleagues first tested their pipeline in a virtual experiment where a floating robotic hand needed to rotate a block into a target position. The digital robot executed the task at a rate of 81 percent accuracy by training on PhysicGen’s massive dataset, a 60 percent improvement from a baseline that only learned from human demonstrations.

The researchers also found that PhysicsGen could improve how virtual robotic arms collaborate to manipulate objects. Their system created extra training data that helped two pairs of robots successfully accomplish tasks as much as 30 percent more often than a purely human-taught baseline.

In an experiment with a pair of real-world robotic arms, the researchers observed similar improvements as the machines teamed up to flip a large box into its designated position. When the robots deviated from the intended trajectory or mishandled the object, they were able to recover mid-task by referencing alternative trajectories from their library of instructional data.

Senior author Russ Tedrake, who is the Toyota Professor of Electrical Engineering and Computer Science, Aeronautics and Astronautics, and Mechanical Engineering at MIT, adds that this imitation-guided data generation technique combines the strengths of human demonstration with the power of robot motion planning algorithms.

“Even a single demonstration from a human can make the motion planning problem much easier,” says Tedrake, who is also a senior vice president of large behavior models at the Toyota Research Institute and CSAIL principal investigator. “In the future, perhaps the foundation models will be able to provide this information, and this type of data generation technique will provide a type of post-training recipe for that model.”

The future of PhysicsGen

Soon, PhysicsGen may be extended to a new frontier: diversifying the tasks a machine can execute.

“We’d like to use PhysicsGen to teach a robot to pour water when it’s only been trained to put away dishes, for example,” says Yang. “Our pipeline doesn’t just generate dynamically feasible motions for familiar tasks; it also has the potential of creating a diverse library of physical interactions that we believe can serve as building blocks for accomplishing entirely new tasks a human hasn’t demonstrated.”

Creating lots of widely applicable training data may eventually help build a foundation model for robots, though MIT researchers caution that this is a somewhat distant goal. The CSAIL-led team is investigating how PhysicsGen can harness vast, unstructured resources — like internet videos — as seeds for simulation. The goal: transform everyday visual content into rich, robot-ready data that could teach machines to perform tasks no one explicitly showed them.

Yang and her colleagues also aim to make PhysicsGen even more useful for robots with diverse shapes and configurations in the future. To make that happen, they plan to leverage datasets with demonstrations of real robots, capturing how robotic joints move instead of human ones.

The researchers also plan to incorporate reinforcement learning, where an AI system learns by trial and error, to make PhysicsGen expand its dataset beyond human-provided examples. They may augment their pipeline with advanced perception techniques to help a robot perceive and interpret their environment visually, allowing the machine to analyze and adapt to the complexities of the physical world.

For now, PhysicsGen shows how AI can help us teach different robots to manipulate objects within the same category, particularly rigid ones. The pipeline may soon help robots find the best ways to handle soft items (like fruits) and deformable ones (like clay), but those interactions aren’t easy to simulate yet.

Yang and Tedrake wrote the paper with two CSAIL colleagues: co-lead author and MIT PhD student Hyung Ju “Terry” Suh SM ’22 and MIT PhD student Bernhard Paus Græsdal. Robotics and AI Institute researchers Tong Zhao ’22, MEng ’23, Tarik Kelestemur, Jiuguang Wang, and Tao Pang PhD ’23 are also authors. Their work was supported by the Robotics and AI Institute and Amazon.

The researchers recently presented their work at the Robotics: Science and Systems conference.

Source link

Up Next

AI coding tools may not speed up every developer, study shows

Don't Miss

ChatGPT analyzed my favorite fictional characters and I was surprised at the insights into my personality – TechRadar

The Editors

Click to comment

AI Research

Axis Communications launches 12 new artificial intelligence dome cameras

Published

24 minutes ago

September 1, 2025

Aurel Dragan

Axis Communications announces the launch of 12 ruggedized indoor and outdoor-ready dome cameras. Based on the ARTPEC-9 chipset, they offer high performance and advanced analytics. AXIS P32 Series offers excellent image quality up to 8 MP. The cameras are equipped with Lightfinder 2.0 and Forensic WDR, ensuring true color and sharp detail even in near total darkness or difficult lighting situations. In addition, OptimizedIR technology enables complete surveillance even in absolute darkness.

Based on the latest platform from Axis, these artificial intelligence dome cameras offer accelerated performance and run advanced analytics applications directly at the edge. For example, they include pre-installed AXIS Object Analytics for detecting, classifying, tracking and counting people, vehicles and vehicle types. They also have AXIS Image Health Analytics, so users receive notifications if an image is blocked, degraded, underexposed or redirected.

Furthermore, some models include an acoustic sensor with AXIS Audio Analytics pre-installed. This alerts users even in the absence of visual cues by detecting shouts, screams or changes in sound level.

Key features:

Outstanding image quality, up to 8 MP;
Indoor and outdoor models;
Variants with pre-installed AXIS Audio Analytics;
Options with different lens types;
Integrated cybersecurity through Axis Edge Vault.

These rugged, vandal- and shock-resistant cameras include both indoor and outdoor versions, with outdoor models operating in an extended temperature range from -40°C to +50°C.
In addition, Axis Edge Vault, the hardware cybersecurity platform, protects the device and provides secure storage and key operations, certified to FIPS 140-3 Level 3.

Source link

AI Research

Breaking Down AI’s Role in Genomics and Polygenic Risk Prediction – with Dan Elton of the National Human Genome Research Institute

Published

1 hour ago

September 1, 2025

The Editors

While protein sequencing efforts have amassed hundreds of millions of protein variants, experimentally determined structures remain exceedingly rare, lagging far behind the number of unresolved structures.

The 2024 UniProt knowledgebase catalogs approximately 246 million unique protein sequences, yet the Worldwide Protein Data Bank holds just over 227,000 experimentally determined three-dimensional structures — covering less than 0.1% of known proteins.

De novo structure elucidation remains a prohibitively expensive and time-intensive endeavor. According to a peer-reviewed article in Bioinformatics, the average cost of X-ray crystallization is estimated at $150,000 per protein.

Even with an annual Protein Data Bank throughput exceeding 200,000 new structures, laboratory workflows struggle to keep pace with the relentless pace of sequence discovery, leaving critical drug targets and novel enzymes structurally uncharacterized.

By harnessing deep learning algorithms to predict three-dimensional conformations from primary sequences, AI-driven models like AlphaFold collapse months of crystallographic work into minutes, directly bridging the gap between sequence abundance and structural insight.

Emerj Editorial Director Matthew DeMello recently spoke with Dan Elton, Staff Scientist at the National Human Genome Research Institute, on the ‘AI in Business’ podcast to discuss how AI is revolutionizing protein structure prediction. Elton concentrates on AI-driven protein engineering and neural-network polygenic risk scoring, outlining a vision for how technology can compress R&D timelines and sharpen disease prediction.

Precision health leaders reading this article will find a clear and concise breakdown of critical takeaways from their conversation in two key areas of AI deployment:

Enhancing polygenic risk stratification: Applying deep learning and neural networks to model nonlinear gene interactions, thereby sharpening disease-risk predictions
Improving rapid structure elucidation: Employing AI-driven protein folding models to predict three-dimensional protein conformations from amino-acid sequences in minutes, slashing timelines for drug discovery and bespoke enzyme engineering

Listen to the full episode below:

Guest: Dr. Dan Elton, Staff Scientist, National Institutes of Health

Expertise: Artificial Intelligence, Deep Learning, Computational Physics

Brief Recognition: Dr. Dan Elton is currently the Staff Scientist at the National Human Genome Research Institute under the National Institutes of Health. Previously, he worked for the Mass General Brigham, where he looked after the deployment and testing of AI systems in the radiology clinic. He earned his Doctorate in Physics in 2016 from Stony Brook University.

Improving Rapid Structure Elucidation

Traditional structural biology methods have long constrained drug discovery and enzyme design workflows. Elton notes that determining a protein’s three-dimensional structure was an extremely difficult problem.

According to Elton, AlphaFold — an artificial intelligence system that predicts the three-dimensional structure of proteins from their amino acid sequences — bypasses these labor-intensive physics simulations by training deep neural architectures on evolutionary and sequence co-variation patterns. It ultimately collapses weeks of bench work into minutes on modern GPU clusters.

Elton explains that open-access folding databases now host over 200 million predicted structures, democratizing discovery by granting small labs the same AI-driven insights previously limited to large pharmaceutical R&D centers.

By collapsing months of laborious X-ray crystallography or NMR experiments into minutes on a modern GPU cluster, companies can now screen thousands of candidate molecules in silico, iterating designs with agility.

Elton emphasizes that this agility not only accelerates lead optimization but also reallocates experimental budgets toward functional assays and ADMET profiling.

Key AI data inputs include:

Amino acid sequences paired with multiple sequence alignments to capture evolutionary constraints
Deep learning models that predict residue-level confidence scores (pLDDT) and contact maps
High-throughput in silico mutagenesis for de novo enzyme design and stability screening

Broadly, integrating AI predictions with targeted experimental workflows has slashed cost-per-structure metrics by orders of magnitude.

This computational acceleration proves particularly valuable for neglected diseases, where the Drugs for Neglected Diseases Initiative now maintains over 20 new chemical entities in its portfolio, partly through AlphaFold-enabled target identification.

DeepMind estimates that AlphaFold has already potentially saved millions of dollars and hundreds of millions of research years, with over two million users across 190 countries accessing the database.

However, Elton’s perspective acknowledges both the revolutionary potential and remaining limitations. While AlphaFold excels at predicting static protein structures, drug development increasingly requires understanding dynamic protein-protein interactions and conformational changes.

The recently released AlphaFold 3 addresses some of these limitations by modeling interactions between proteins and other molecules, including RNA, DNA, and ligands. Google claims in an interview with PharmaVoice that there was at least a 50% improvement over existing prediction methods for protein interactions.

Enhancing Polygenic Risk Stratification

Building on these structural breakthroughs, Elton next turns from folded proteins to the genome itself, where AI is poised to redefine risk prediction and gene-editing delivery.

Conventional polygenic risk-score frameworks rely on additive, linear regression models that perform well for highly heritable traits like height but fail to capture complex gene–gene interactions.

Elton explains that the way genes are associated with phenotypes is not simply linear. Nonlinearities exist as well, highlighting the limitations of sparse linear predictors.

Neural network and deep learning architectures offer a path to uncover epistatic effects, yet Elton cautions that such models demand unprecedented data and compute scales. He notes that to predict a condition like autism or even intelligence, researchers would need between 300,000 and 700,000 sequences, necessitating tens of trillions of letters or tokens.

In other words, matching the data scale of GPT-4 becomes a prerequisite — demanding robust cohort assembly, cross-biobank harmonization, and petascale compute infrastructure.

Elton candidly notes that the added value of using a neural net or a language model actually might be relatively small for some traits where linear models already capture most genetic effects. For heritable characteristics like height, for example, the added neural net value is relatively small because linear predictors explain all the heritability.

This honest assessment reflects the understanding required to prioritize which genetic traits and clinical applications justify the massive computational investment needed for neural network-based polygenic prediction.

Elton also warns that handling tens of trillions of tokens per project requires more than raw compute; it mandates rigorous data-management frameworks that ensure privacy, regulatory compliance, and security. Cloud architects and life-science IT leaders should therefore adopt:

Encryption-at-rest
Role-based access control
Immutable audit trails to safeguard personally identifiable information

Beyond prediction, Elton mentions that AI is also transforming precision gene editing workflows. Elton describes ex vivo therapies — when blood is extracted, treated with genetic editing, and ultimately returned into the bloodstream.

In this way, AI tools can now fine-tune viral shells so they target the right tissues and optimize guide-RNA instructions to avoid accidental gene cuts.

Source link

AI Research

Strategies for CPAs to Become Artificial Intelligence (AI) Savvy

Published

1 hour ago

September 1, 2025

Robbie Bishop-Monroe, DBA, CPA, CFE, Lesia Quamina, JD, LLM, CPA (inactive)

Of the many kinds of technologies that professionals have encountered in recent years, artificial intelligence (AI) presents perhaps the greatest challenge. CPAs that do not become comfortable with AI and integrate it into their toolkit risk falling behind the technology curve. This article aims to demystify the AI concept for accountants and provide useful ways that CPAs and their organizations can use AI tools. This article also shares useful resources for those seeking to become AI savvy.

***

Artificial intelligence (AI) has become prominent in business operations, company investments, budgets, and strategic plans, with corporate investment of approximately $252.3 billion in 2024(Artificial “Intelligence Index Report 2025,” Stanford University, https://tinyurl.com/23ssm6y9). The promise of AI is based on machines taking over activities once done by the human brain. With the release of AI technologies like ChatGPT, Copilot, Bard, and Dall-E seemingly making good on that promise as they introduce to society how AI could amplify human effectiveness, increase innovation, improve business efficiencies, and enhance customer service. As a result, many businesses have begun utilizing AI as part of business operations in content management, cybersecurity, fraud management, and customer support. According to Accenture (Reilly, et al., “AI: Built to Scale,” Accenture, November 2019, https://tinyurl.com/353pzsxp), an astounding 84% of business executives believe they need to use AI for business growth and to assist in achieving their strategic goals; however, 76% experience challenges with how to adapt and adopt AI effectively into their business practices. The impact of AI is not limited to large businesses. A United States Chamber of Commerce report found that one in four small businesses use AI to help enhance marketing and communications performance [“Empowering Small Business: The Impact of Technology on U.S. Small Business (Second Edition),” U.S. Chamber of Commerce, September 2023, https://tinyurl.com/yc4ut2eh]. While AI is increasingly an integral part of the business environment, integrating it into everyday practice remains a challenge.

AI and the Accounting Profession

Proponents predict that the accounting profession will leverage AI in several ways: 1) automating routine tasks such as data entry and transaction processing; 2) conducting financial analysis and forecasting due to its ability to process large financial datasets; 3) automating audit functions by analyzing financial transactions in real-time to identify discrepancies and generate reports; 4) improving tax compliance and planning by automating tax calculations and providing optimal tax saving opportunities and strategies (Jason Ackerman, “Artificial Intelligence May Be Coming Sooner than Expected,” The CPA Journal, May/June 2023, https://tinyurl.com/5ytxsryr).

The Big Four—Pricewaterhouse-Coopers (PWC), KPMG, Deloitte, and Ernst & Young (EY), have invested heavily in AI. In audit practice, the firms have developed AI tools to automate management of the audit process via systems like PWC Halo (“Audit of General Ledger with Halo,” PwC, https://tinyurl.com/3vrwejdz), KPMG Clara [“Bringing Clarity to the Audit with AI (Artificial Intelligence),” KPMG, https://tinyurl.com/3d5yea9u], EY Canvas (D. D’Egidio, et al., “Our Global Audit Platform, Powering Our One Global Audit, Is at the Heart of Our Digital Audit Offering,” EY, https://tinyurl.com/yj32zhm8) and Deloitte Omnia (Schmidt, et al., “Audit Innovation,” Deloitte, https://tinyurl.com/mr347vx3).

Additional AI tools that aid in fraud detection are PWC’s GL.ai (G. Rapsey, et al., “Harnessing the Power of AI to Transform the Detection of Fraud and Error,” PWC, https://tinyurl.com/3zszj8kn), and EY’s EY Helix (D. D’Edigo, et al., “EY Helix,” EY, https://tinyurl.com/mst26c4r). Both AIs are embedded on client platforms and can review billions of data points in milli-seconds and analyze the data to detect anomalies in the general ledger. Deloitte also has an array of AI tools (C. Oh, “Deloitte Drives the Power and Potential of Advanced AI and Generative AI to Internal Audit,” Deloitte Press Release, August 2024, https://tinyurl.com/3mjetx8x) including Argus, one of Deloitte’s oldest AI tools, which extracts accounting information from any type of electronic document to allow auditors to examine a large sample (T. H. Davenport, “The Power of Advanced Audit Analytics: Everywhere Analytics,” Deloitte, 2016).

The Big Four AI tools mentioned above are not comprehensive, but it should provide a roadmap for what is next as the firms continue investing in AI. In 2023, PWC indicated that they plan to invest approximately $1 billion over three years to train existing staff in AI, hire new AI staff, integrate AI platforms into their business operations, and provide consulting services for companies on how to incorporate AI into their business practices (A. Loten, “PricewaterhouseCoopers to Pour $1 Billion Into Generative AI,” Wall Street Journal, April 2023, https://tinyurl.com/2fh2u2p5). During the same period, KPMG announced an investment of $2 billion in AI and cloud services to enhance and automate their consulting, audit, and tax services to enable staff to utilize more time on providing advice to clients (M. Mauer, “KPMG Plans $2 Billion Investment in AI and Cloud Services,” Wall Street Journal, July 2023, https://tinyurl.com/3pef2f3e).

Similarly, EY invested $1.4 billion to launch EY.ai, which assists companies with an AI platform to perform business operations more efficiently in strategy, transactions, transformation, risk, insurance, and tax. EY has also invested in cloud and automation technologies. Even though EY had an existing platform, EY Fabric, this additional investment signals the significance of its investment in AI (“EY Launches Artificial Intelligence Platform ‘EY.ai’ and Invests US$1.4 Billion,” EY Press Release, September 2023, https://tinyurl.com/y3yen6ep). Finally, in April of 2024, Deloitte announced that they plan to invest $2 billion in Industry Advantage to provide industry-focused solutions to clients, including AI and cybersecurity (Deloitte Press Release, https://tinyurl.com/yncx3r44).

Fortunately, AI is not just for the Big Four, and small firms or individuals are not required to invest millions or billions. This article highlights AI tools that can be useful to CPAs and provides information for professionals just starting their AI journey.

What Is AI?

According to Accenture, “Artificial intelligence is a constellation of many different technologies working together to enable machines to sense, comprehend, act, and learn with human-like levels of intelligence” (“Artificial Intelligence,” Accenture, https://tinyurl.com/2mbcnhcw). This could include machine learning (“AI 101: What Is Machine Learning?” Accenture, https://tinyurl.com/2bcrpjpw) and natural language processing (“The Basics of Natural Language Processing,” Avathon, https://tinyurl.com/4cdesfws). According to a report by Forbes Advisor, the most common uses of AI are as digital assistants, chatbots, and machine learning (K. Haan, “How Businesses Are Using Artificial Intelligence,” April 2024, https://tinyurl.com/mzr2hj9d). This article will refer to AI in the broadest and most common usage.

As the use of AI has spread, many have questioned whether it could replace human accountants. According to KPMG’s Cliff Justice, AI “won’t replace the human at the human-to-human interactions level … AI tools are really good at pulling out information and making predictive choices, but they can’t replace human judgment” (S.J. Steinhardt, “Big Four Agree: AI Will Not Replace Accountants,” The Trusted Professional, August 2023, https://tinyurl.com/nzthbue7).

AI can make business processes more efficient and provide some technical support; however, only a live, skilled accountant can provide in-depth analysis and perspective as well as confirm that the information generated is correct (“Will AI or Automation Replace Accountants? A Critical Look at What the Future Holds,” Financial Cents, https://tinyurl.com/55epkkpf). Nevertheless, it is critical for CPAs to adapt and adopt AI skills to stay marketable and enhance business efficiency.

The ABCs of AI

The authors conducted an informal survey of smaller CPA firms about whether they used AI in their practice and their common responses were “No” or “AI?”. Thus, before delving into practical ways to integrate AI into an accounting practice, it’s helpful to review the phases of AI adoption and introduce some AI hands-on tools.

Assess one’s position on the AI journey.

Assessing where one falls in the AI journey is critical to determine what comes next. Shown in Exhibit 1, the AI Journey Roadmap from the RSM Play-book (Beyer, et al., “Here Is The Middle Market Artificial Intelligence (AI) Playbook,” RSM, https://tinyurl.com/mtepvftf), outlines five different phases of the AI adoption and implementation journey: Phase 1, AI education; Phase 2, AI strategy and assessment; Phase 3, AI preparation; Phase 4, AI execution; and Phase 5, AI support and maintenance. Most smaller CPA firms and sole practitioners may be in the AI education phase, but their goals may include AI playing a greater role down the road. On the other hand, larger firms could be between AI execution (Phase 4) and AI support and maintenance (Phase 5). (As a larger firm, RSM touts the wide variety of AI consulting services it provides to clients, https://tinyurl.com/cbvabamx.) It is important to note that although a firm could be in the AI support and maintenance phase, AI education is an iterative process.

EXHIBIT 1

Become familiar with basic artificial intelligence (AI) tools.

CPAs may want hands-on experience with AI to demystify the subject. For example, one could test a sample of basic online AI tools to determine which is preferable. Two common options are Gemini, powered by Google, and ChatGPT [S. Ortiz, “What Is Google’s Gemini AI tool (formerly Bard)? Everything You Need to Know,” ZDNet, February 2024, https://tinyurl.com/3av2fwsz]. The primary difference between the AI tools is that ChatGPT Premium and Enterprise use Microsoft Bing data and are no longer restricted to information before 2021 as previously (A. Pequeno, “Major ChatGPT Update: AI Program No Longer Restricted to Sept. 2021 Knowledge Cutoff After Internet Browser Revamp,” Forbes, September 2023, https://tinyurl.com/58664w8a), whereas Gemini uses Google data.

To test how each AI tool works, one of the authors signed up for both. First, I signed up for Gemini (https://tinyurl.com/2wnvdc45), agreed to the terms and conditions, and when the prompt screen was displayed, I typed “How can accountants use Artificial Intelligence?” The results (Exhibit 2) included automating tasks, data analysis, auditing, client communication, and budgeting. The caveat is that a disclaimer states, “Gemini may display inaccurate data.” This is highlighted because it is important to note that it is critical to verify the information provided by ChatGPT or Gemini.

EXHIBIT 2

Next, I signed up for ChatGPT (https://tinyurl.com/4hp4c9tm) and typed in “How can accountants use artificial intelligence?” Interestingly, ChatGPT states, “Don’t share sensitive info. Chats may be reviewed and used to train our models.” In the author’s experience, ChatGPT provided more content than Gemini. For example, ChatGPT shared eight ways accountants can use AI (Exhibit 3), whereas Gemini gave only five.

EXHIBIT 3

These tools are continuously being enhanced; for example, PWC formed a strategic alliance to enhance AI tools, including Gemini (C. Sedlak, “PwC and Google Cloud Announce Strategic Collaboration to Accelerate Enterprise Adoption of Vertex AI and Gemini Models,” PWC, April 2024, https://tinyurl.com/h684v7bv).

Practical AI Strategies—Where to Start?

Once one knows where they are in the AI journey and some of the available tools, it is time to explore practical ways to implement AI.

Keeping up to date with accounting trends and regulations.

Staying current with trends in AI is crucial. For example, Charles P. Myrick, a CPA firm owner in Washington, DC (https://tinyurl.com/7kx2sard), uses AI to research accounting trends and regulations. He emphasizes the importance of conducting additional research and analysis, as AI cannot be fully relied upon for accuracy. Even though AI tools are being utilized more than ever before, users should confirm that their information is based on the most recent available data instead of blindly relying on the AI query. For example, users can upload a newly issued Accounting Standards Update to their preferred AI tool and then query the AI to obtain a summary of the guidance.

Writing and researching information for reports or documents.

Writing and researching are an important part of business operations. Whether drafting emails, creating business memos, preparing reports for clients, or developing policies and procedures manuals, AI can significantly improve efficiency. For example, AI can be helpful for repetitive business tasks such as writing emails to clients (B. Oliver, “How Artificial Intelligence Can Help Save Accounting,” Journal of Accountancy, November 2023, https://tinyurl.com/y4v2mcn4) and reduce the time to respond to inquiries. In Outlook or Gmail, for example, the predictive text feature of AI automatically suggests phrases that can be written next with a swipe or pressing the tab, speeding up email communication and enhancing customer service. Grammarly (https://www.grammarly.com) is another tool that can assist in editing e-mails, Word documents, or PowerPoint presentations.

According to the Journal of Accountancy, writing a research memo can take up to “40 to 80 hours” (Oliver 2023). For example, tools like ChatGPT or Gemini can be used to query a specific topic. Although the information gathered would need to be verified, AI can provide a starting point. AI can also summarize large datasets, such as government regulations or research articles, saving time and resources.

Additionally, AI could aid business development by efficiently responding to Requests for Proposals (RFP) questions (Beyer). AI tools in this area include autorfp (https://autorfp.ai) and DataRobot (https://tinyurl.com/53jj-dv58). For example, if a company is considering submitting an RFP, AI tools have the capability to review data from its website and previous RFP content, then summarize information, and prepare responses to the RFP questions. This can increase efficiency and reduce operational costs.

Summarizing meeting notes.

Harvard Business Review estimates executives spend an average of 23 hours each week in meetings (L. Perlow, et al., “Stop the Meeting Madness,” Harvard Business Review, August 2017, https://tinyurl.com/bddnzth5). With AI, there is no longer a need to manually type meeting summaries from a voice recording. AI can assist with repetitive tasks by automatically delivering a written summary of the meeting minutes. Notetaking AIs such as Circleback (https://circleback.ai), Krisp (https://krisp.ai), and Granola (https://www.granola.so) offer seamless note-taking solutions, ensuring that all attendees are on the same page. These tools range from in-meeting bots to external notetakers (M. Peng, “The Best AI Note-Taking Tools for Meetings,” Charter, June 2024, https://tinyurl.com/4nfd5vja). Another example is the AI-powered Zoom (https://zoom.us), which can summarize the conversation during a meeting. Latecomers to the Zoom meeting can catch up by chatting with the AI assistant about what they missed. The Zoom AI assistant provides documented summaries in multiple languages (https://tinyurl.com/238m2csc). Whether for routine meetings or client discussions, having AI produce automatic notes documenting and summarizing saves time and money and automates routine tasks. The caveat is that, as with human notetakers, a summary must be reviewed for accuracy.

Enhancing customer service with chatbots.

According to a survey by Forbes Advisor, the most popular use of AI tools is customer service (https://tinyurl.com/yc5z7xv9). This trend is led by chatbots, which use AI and natural language processing to interact via text and voice with users. They are available 24 hours a day and can respond to common customer questions. Website service companies that cater to CPAs, such as Get Net Set (https://getnetset.com) or CPA Site Solutions (https://www.cpasitesolutions.com), can incorporate AI into a firm’s website. Another firm, PJC Group, LLC, uses AI through a chatbot to assist website visitors with questions (Exhibit 4).

EXHIBIT 4

Streamlining auditing, book-keeping, and consulting services.

Ackerman (2023) predicted that AI would impact accounting, particularly by automating repetitive tasks. AI-powered tools like optical character recognition (OCR) technology are already making a difference. For example, QuickBooks Booke AI can scan and extract data from invoices, bank statements, receipts, and other financial documents or reports, categorizing and summarizing them to assist with bookkeeping and reconciliations. It also provides a tool for fixing bookkeeping transactions and conducting reconciliations (https://tinyurl.com/72uaa5wj). This automation not only saves time but also improves accuracy by reducing errors. Excel users may consider using the DataSnipper (https://www.datasnipper.com) add-in, which integrates AI OCR to scan invoices, bills, or other documents into Excel. This provides a direct link to the source document, creating a clear audit trail when confirming the accuracy of financial data.

AI is also playing a role in detecting fraudulent transactions (https://tinyurl.com/cruyfemr), which can also help CPA firms delivering consulting services or performing an audit to conduct analytics to identify fraudulent transactions. AI tools such as Mind-Bridge (https://www.mindbridge.ai) can efficiently conduct an overview of the company’s financial data, determine risk areas, identify anomalies, and provide the auditor with a list of recommended areas to audit.

Professional branding and marketing.

In a media landscape driven by social media, building a personal brand is essential to stand out from the pack. AI tools can help create content, analyze customer engagement, and foster community. Platforms like Turbo Logo (https://turbologo.com), Looka (https://looka.com), and Canva (https://www.canva.com) can create eye-catching logos. AI tools such as Tugan.ai (https://www.tugan.ai) and Writesonic (https://writesonic.com) can help CPAs create content derived from marketing material already on the firm’s website, such as sales emails or newsletters.

Given the critical role of social media, an organization’s online presence is key to engaging users and building community. To this end, tools such as Social Pilot (https://www.socialpilot.co), Social Bee (https://socialbee.com), and Hootsuite (https://www.hootsuite.com) can help firms manage social media accounts through design, scheduling, and engagement services.

Another budget-friendly way to use AI in personal branding is through professional headshots for company websites or LinkedIn. AI-powered tools can do photo editing, background removal, virtual photo studios, and AI-generated headshots. For example, HeadshotsPro (https://www.headshotpro.com), AI Suitup (https://www.aisuitup.com), and Secta.ai (https://secta.ai) use AI to convert personal selfies into usable professional headshots.

Conducting firm training.

AI provides an opportunity to upskill sole proprietors, partners, managers, and staff across all levels. Training can be conducted by utilizing AI to query for solutions to an issue encountered during an engagement. For example, Deloitte employs an AI called DARTbot (Will Bible, “Generative AI in Accounting: Opportunities and Risks to Assess Today,” Deloitte, https://tinyurl.com/56ck4h96) to support its employees with daily tasks, answering accounting questions, and conducting research, enabling employees to focus on higher-level workflow. For smaller firms, AI tools such as Synthesia (https://www.synthesia.io), and InVideo (https://invideo.io) can transform textual training lessons into engaging videos. Employees can access the material on-demand, and it can be tailored to individual skill level. In addition, chatbots can provide on-going support to employees by answering routine questions or guiding them to available resources.

Getting Prepared

Although there have been large investments in AI, according to a PWC survey, 88% of companies struggle to realize quantifiable value from AI (“PwC Pulse Survey: Focused on Reinvention,” PwC, https://tinyurl.com/yc4ckcz2). AI promises to be a critical element in business strategy, today and into the future. Integrating AI is an enormous task, and CPAs would do well to solicit the assistance of consultants that can guide an organization through the process. Exhibit 5 presents a list of technology skills, resources, and pricing that can help CPAs beginning their AI journey.