The IRS will deploy AI to increase audits on the wealthy; not everyone on Capitol Hill is happy about it

Welcome to The Dispatch! We are the newsletter that keeps you informed about AI. Each weekday, we scour the web to aggregate the many stories related to artificial intelligence; we pass along the news, useful resources, tools or services, technical analysis and exciting developments in open source. Even if you aren’t an engineer, we’ll keep you in touch with what’s going on under the hood in AI.

IRS to leverage AI in sweeping effort to restore fairness to tax system

The story: The Internal Revenue Service has announced a massive new effort to increase enforcement and audits on wealthy taxpayers and large corporations, while reducing burden on lower- and middle-income taxpayers. This effort will be powered by the $80 billion in funding from the Inflation Reduction Act (signed into law in late 2022) - and artificial intelligence. The technology allows for increased scrutiny upon what has been, historically, a very difficult segment to audit - corporate partnerships:

“With the help of AI, the selection of these (large and complex partnership) returns is the result of groundbreaking collaboration among experts in data science and tax enforcement, who have been working side-by-side to apply cutting-edge machine learning technology to identify potential compliance risk in the areas of partnership tax, general income tax and accounting, and international tax in a taxpayer segment that historically has been subject to limited examination coverage.”

More Details:

• Priority areas include cryptocurrency non-compliance, foreign bank accounts, large partnership balance sheet discrepancies, and labor broker schemes used to avoid payroll taxes. The IRS has also noted there are over 1,600 millionaires who owe at least $250,000 in back taxes.

• AI helps process the massive amounts of data required for audits much faster than manual analysis would allow. This helps all parties get a better picture of the ecosystem and develop critical reports to governmental bodies.

• Audit rates will not increase for those earning under $400k. The IRS wants to avoid unnecessary audits for average taxpayers.

• The IRS will also focus on protecting low-income taxpayers from predatory tax preparers improperly claiming credits.

Takeaways: This New York Times op-ed from a governance fellow at the Brookings Institution back in 2016 highlights public sentiment about tax evasion:

“Asked what bothers them most about taxes, Americans overwhelmingly say the feeling that the wealthy and corporations are not paying their fair share. This is the top issue for nearly two-thirds of Americans. In contrast, 8 percent of Americans say that their biggest concern is the amount they personally pay in taxes. What upsets most people about taxes is not the amount they contribute. They are angry about the amount that the wealthy can avoid contributing.”

It took little time for a direct response from Capitol Hill, and Congressman Clay Higgins (R-LA) has sponsored extreme legislation over the announcement - claiming that ‘Democrats love AI’ and are ‘weaponizing’ it, implying that the technology is a dividing line between parties.

It has not been, so far. Perhaps the issue of using AI to go after wealthy tax evaders and unscrupulous corporations will tip that scale.

However, Sunak faces challenges on both the world stage and at home. President Biden's announced absence already deals a blow, and wavering commitment from major European leaders also jeopardizes the event's stature. The recent G20 underscores the difficulty of choreographing a compelling global summit, and questions linger about the UK's convening power post-Brexit.

However, AI avatars prompt many ethical concerns, especially regarding misuse, authenticity in relationships, and gender bias. As these avatars become more immersive with advancements in areas like AR and VR, careful development and thoughtful implementation becomes more important. AI avatars have many promising applications - but the technology can’t replicate genuine human experiences and relationships.

• (Excerpt from Elon Musk’s biography, published today) How Elon Musk set Tesla on a new course for self-driving

• Meta sets GPT-4 as the bar for its next AI model, says a new report

• Roblox’s new AI chatbot will help you build virtual worlds

• Tim Burton is not amused by AI recreations of his style: ‘It’s like a robot taking your humanity’

• Elusive Ernie: China's new chatbot has a censorship problem

• Nvidia's dominance in AI chips deters funding for startups

DATA replaces Siri with ChatGPT, allowing users to control their iPhone and apps via conversational voice commands processed by AI. Users can create an account on HeyDATA.org, connect their OpenAI API key, and install Shortcuts for features like long-term memory, calendar and weather integration, summarizing emails, generating visual guides, and transcription.

To investigate, the researchers tested LLMs of varying sizes on out-of-context reasoning tasks. They found that larger models performed better on these tasks, suggesting a correlation between scale and the potential emergence of situational awareness. While current LLMs are still far from fully aware, the study provides an initial framework to monitor this ability. More refined experiments will be needed to accurately predict and control the development of situational awareness as AI models advance.

• (Open source) ChatDev: communicative agents for software development

• Nvidia TensorRT-LLM supercharges Large Language Model inference on H100 GPUs

• Doppelgangers: learning to disambiguate images of similar structures