AI is forcing corporate boards to change how they operate, with the most aggressive companies appointing AI bots as observers to their boards and putting tech at the center of their board strategy work.

Why it matters: The world's largest companies are increasingly obsessed with AI — mentioning it repeatedly in 2024 earnings calls — but most boards lack the expertise to effectively guide AI strategies.

• New uses of AI could help boards make more informed decisions.

The big picture: The pace of AI innovation is pushing executives to develop AI strategies quicker than they have for previous new technologies, and boards are now part of that wave of change.

Case in point: One of the UAE's most valuable public companies, International Holding Company, has appointed Aiden Insight to its board as an "AI observer."

• Aiden Insight is the persona of a tool called BoardNavigator, created by G42 — the Gulf region AI company that recently obtained a $1.5 billion investment from Microsoft.
• BoardNavigator was built with Microsoft's Azure OpenAI service, in cooperation with Microsoft, G42 said in a statement.

How it works: G42 promises that Aiden Insight will provide "real-time insights to inform discussions and guide decisions" during business meetings, which it says is possible because of continuous data analysis and ethical and compliance monitoring.

• The tool combines a company's own data with external market trend data to offer advice, and G42 says it works best for energy, health, finance and technology companies.

Context: Non-AI software has been changing board governance for years — most noticeably through "board portals" offered by companies such as Nasdaq, which create a secure environment for boards to deliberate and take action.

Driving the news: The National Association of Corporate Directors created a 24-member commission April 16 to develop AI principles to guide its 23,000 or so members who sit on boards.

By the numbers: Only 13% of S&P 500 companies have at least one director with AI-related expertise, according to an analysis of SEC filings by ISS-Corporate, a governance consultancy.

• Just eight companies in the S&P 500 have formal AI board oversight systems, per ISS, and only around 1 in 7 Fortune 500 boards have a broader science or technology committee.

What they're saying: "The use of AI in the boardroom is a game changer," Kiril Evtimov, group chief technology officer for G42, said in a statement.

• "CEOs sometimes don't understand the issue well enough to organize the different interests inside the organization" around AI, and boards need AI competence to spot that, Friso van der Oord, senior vice president at NACD, tells Axios.
• "It's crazy for a board to [still] get updates every 90 days with PDFs when predictive AI can spot a revenue issue two weeks into a quarter," Steve Singh, managing director of Madrona Venture Group, tells Axios.
• Singh, who is also a board member at Clari, a revenue management platform used by executives and boards, says: "Soon with generative AI, agents will be working on the board's behalf, prodding a company's revenue leaders."

Between the lines: Experts suggest that, thanks to AI and other tech-driven trends, executives need to hear from their boards more frequently, rather than once a month, or every three months.

Yes, but: All this acceleration of board work could blur the lines between a company's operational management and board oversight.

• Seating an AI bot on a board also raises novel questions about liability for the AI's actions. Alissa Kole, managing director of advisory firm Govern Center, argues that responsibility lies with the AI's developers, "effectively upending the entire concept of board liability insurance."

Flashback: 45% of respondents to a 2015 World Economic Forum technology survey predicted the first AI board directors would appear by 2025.

Adobe is bringing text-to-image capabilities directly into Photoshop today and releasing an improved version of its AI-powered Firefly image engine, Ina reports.

Why it matters: The company is trying to maintain a tricky balance of assuaging content creators' concerns about the technology while also offering AI tools to those who want them.

Driving the news: The new version of Photoshop will allow people to type a description of an image or background they want, which is then added as a new layer.

• Adobe says the new version of Firefly renders details that past models have struggled with, including straight lines and fingers.
• Like its predecessors, version 3 of Adobe Firefly Image is designed to be "commercially safe," meaning it was trained on data that is either in the public domain or licensed by Adobe.

Yes, but: Adobe has recently come under some criticism for training Firefly with some stock images that were created using other AI tools, such as Midjourney.

What they're saying: Although many people launch Photoshop to edit an existing document, others want to start a new project from scratch or add a background to a single image, such as a product.

• "We see some customers getting lost in that initial step," Adobe VP of generative AI Alexandru Costin tells Axios. "So we're introducing 'generate image' right in the workflow not only for when you have a new document, but when you want to add a new layer to a document and create the scene."

Between the lines: Adobe gives Photoshop users the ability to use a reference image in creating the AI image. That allows companies to match their brand style and other assets.

• However, its legal protection for Firefly only extends if users don't introduce copyright infringement through this reference image.

Microsoft today began publicly sharing Phi-3, an update to its small language model that it says is capable of handling many tasks that had been thought to require far larger models, Ina reports.

Why it matters: Smaller models use less computing power than larger ones and, in many cases, can run on smartphones or laptops, offering additional performance and privacy benefits.

Driving the news: Microsoft is adding Phi-3 to its own Azure model gallery as well as releasing it on open source model site Hugging Face and Ollama, a platform designed to help people run models on their own machines.

• Microsoft is starting with Phi-3 mini, a version of the model trained on a smaller amount of data (3.8 billion parameters).
• Two other models — still considered lightweight by today's standards — are coming shortly: Phi-3 small is trained on 7 billion parameters and the largest, Phi-3 medium, is trained on 14 billion parameters.
• The company says Phi-3 can outperform similar-size models and even slightly larger ones, thanks in part to the high-quality data on which it is trained.

What they're saying: Microsoft says Phi-3 isn't designed to replace large language models, but can work in places large models don't, including running on devices.

• "If you have a very, very high-stakes application, let's say in a health care scenario, then I definitely think that you should go with the frontier model: the best, most capable, most reliable," Microsoft VP Sébastien Bubeck tells Axios.
• For other uses, other factors matter more, including speed and cost. "That's where you want to go with Phi-3," Bubeck says.

• Google fired about 20 additional workers over protests of a cloud contract with the Israeli government, according to an employee activist group. (Axios)
• Meta is opening up its Quest headset software, creating a platform — Horizon OS — in hopes of generating more specific uses for the device. (ArsTechnica)
• Trading places: Joe Tyrrell, CEO of customer experience company Medallia, has left the company. (Axios)
• Arkansas-based Xtremis is looking to build a marketplace for mobile carriers and the military to rent or lease radio spectrum, which carries your phone's data. (Axios)

Check out this mind-bending street art.