Axios AI+

A floating, translucent blue 3D render of the human brain.

January 16, 2024

Hi, it's Ryan, reporting to you from the Swiss Alps, where the World Economic Forum is underway. The WEF is leaning into the metaverse and Axios is leaning hard into AI at Axios House. Today's AI+ is 1,300 words, a 5-minute read.

1 big thing: AI doesn't forget, and that's a problem

Illustration: Aïda Amer/Axios

Users want answers from artificial intelligence, but sometimes they want AI to forget things, too — creating a new category of research known as "machine unlearning," Axios' Alison Snyder reports.

Why it matters: Interest in techniques that can remove traces of data without degrading AI models' performance is driven in part by copyright and "right to be forgotten" laws, but also by concerns about biased or toxic AI outputs rooted in training data.

Deleting information from computer storage is a straightforward process, but today's AI doesn't copy information into memory — it trains neural networks to recognize and then reproduce relationships among bits of data.

  • "Unlearning isn't as straightforward as learning," Microsoft researchers recently wrote. It's like "trying to remove specific ingredients from a baked cake — it seems nearly impossible."

Driving the news: A machine unlearning competition that wrapped up in December asked participants to remove some facial images used to train an AI model that can predict someone's age from an image.

  • About 1,200 teams entered the challenge, devising and submitting new unlearning algorithms, says co-organizer Peter Triantafillou, a professor of data science at the University of Warwick. The work will be described in a future paper.

What's happening: Researchers are trying a variety of approaches to machine unlearning.

  • One involves splitting up the original training dataset for an AI model and using each subset of data to train many smaller models that are then aggregated to form a final model. If some data then needs to be removed, only one of the smaller models has to be retrained. That can work for simpler models but may hurt the performance of larger ones.
  • Another technique involves tweaking the neural network to de-emphasize the data that's supposed to be "forgotten" and amplify the rest of the data that remains.
  • Other researchers are trying to determine where specific information is stored in a model and then edit the model to remove it.
  • One obvious way to remove the influence of a specific piece of data is to take it out of the training data and then retrain the model, but the high cost of computation means that is basically a "non-starter," says Seth Neel, a computer scientist and professor at Harvard Business School.

Yes, but: "Here's the problem: Facts don't exist in a localized or atomized manner inside of a model," says Zachary Lipton, a machine learning researcher and professor at Carnegie Mellon University. "It isn't a repository where all the facts are cataloged."

  • And a part of a model involved in knowing about one thing is also involved in knowing about other things.

Zoom in: There's particular interest in unlearning for generative language models like those that power ChatGPT and other AI tools.

  • Microsoft researchers recently reported being able to make Llama 2, a model trained by Meta, forget what it knows about the world of Harry Potter.
  • But other researchers audited the unlearned model and found that, by rewording the questions they posed, they could get it to show it still "knew" some things about Harry Potter.

Where it stands: The field is "a little messy right now because people don't have good answers to some questions," including how to measure whether something has been removed, says Gautam Kamath, a computer scientist and professor at the University of Waterloo.

  • It's a pressing question if companies are going to be held liable for people's requests that their information be deleted or if policymakers are going to mandate unlearning.
  • Neel says, "For simple models, we know how to do unlearning and have rigorous guarantees," but for more complex models there isn't "consensus on a single best method and there may never be."

What to watch: For low stakes problems it might be sufficient to stop a model from reproducing something verbatim, but serious privacy and security issues might require complete unlearning of information.

  • Here, Lipton says, near-term policy mandates should "proceed under the working assumption that (as of yet) mature unlearning technology does not exist."

2. Execs admit they're not ready for AI wave

Animated illustration of a robot tightening and adjusting its necktie, followed by its teeth sparkling.

Illustration: Brendan Lynch/Axios

Top executives are admitting they're far from ready to deal with changes wrought by generative AI, more than a year after ChatGPT exploded on the business scene, according to a new global survey by Deloitte's AI institute.

Why it matters: The problems may only get worse. Executives who reported the most investment and knowledge in generative AI capabilities are the ones most worried about the technology's impact on their businesses.

Details: The survey of 2,800 director to C-suite level executives— released as the World Economic Forum in Davos kicks off — found that only 1 in 5 executives believes their organization is "highly" or "very highly" prepared to address AI skills needs in their company.

  • Just 1 in 4 believe their organizations are well-prepared to address AI governance and risks.
  • Only 47% say they are sufficiently educating employees about AI.

The majority of executives said their organizations were focused on the tactical benefits of AI, such as improving efficiency and cost reduction, rather than using it to create new types of growth.

What they're saying: "If you're going to look at this as some side initiative — a department of generative AI that's going to come up with all the use cases — it's going to be a massive failure," Joe Ucuzoglu, Deloitte Global CEO, tells Axios in an exclusive interview.

  • "There's a talent scarcity in the key areas it takes to activate all of this, so it's impossible for any organization to have all the expertise within its own four walls," Ucuzoglu said.

3. Microsoft expands Copilot subscriptions

Image: Microsoft

Microsoft on Monday announced new subscription AI services aimed at individuals and small businesses that will include, among other things, the ability to create custom chatbots, Ina reports.

Why it matters: The move allows access to tools that Microsoft had previously reserved for larger organizations.

Details: A $20-per-month Copilot Pro subscription includes access to Copilot in Office apps like Word, Excel, PowerPoint, Outlook and OneNote. (This had previously been available only to large businesses with a Microsoft 365 subscription.)

  • Custom chatbots, which Microsoft is calling Copilot GPTs, are "coming soon."
  • The subscription also includes faster text-to-image creation as well as higher image quality and the option for images to be created in landscape image format. Subscribers also get priority access to updated models, starting today with OpenAI's GPT-4 Turbo.

Yes, but: The move opens up yet another area in which Microsoft and OpenAI are offering similar products targeting similar audiences.

  • OpenAI's ChatGPT Plus offers some of the same features, including the ability to create custom chatbots, at a similar price.
  • Microsoft says it expects that its custom Copilot GPTs will be interoperable with OpenAI's at some point.

Meanwhile: Microsoft is also making its Copilot for Microsoft 365 available to smaller businesses for $30 per person per month. (It had previously been limited to businesses willing to buy 300 or more subscriptions.)

4. Training data

5. + This

Don't lazily order a bottle of water through room service in Davos — one AI+ reader reports it'll set you back $50 with the tip and "service charge" (what's the difference, you may ask) at four-star hotels here.

Thanks to Scott Rosenberg and Megan Morrone for editing this newsletter.