While supercomputers—most famously IBM’s Deep Blue—have long surpassed the world’s best human chess players, generative AI still lags behind due to their underlying programming parameters. Technically speaking, none of the current generative AI models are computationally capable of beating dedicated chess engines. These AI don’t “know” this, however, and will continue chipping away at possible solutions—apparently with problematic results.

To learn more, the team from Palisade Research tasked OpenAI’s o1-preview model, DeepSeek R1, and multiple other similar programs with playing games of chess against Stockfish, one of the world’s most advanced chess engines. In order to understand the generative AI’s reasoning during each match, the team also provided a “scratchpad,” allowing the AI to convey its thought processes through text. They then watched and recorded hundreds of chess matches between generative AI and Stockfish.

The results were somewhat troubling. While earlier models like OpenAI’s GPT-4o and Anthropic’s Claude Sonnet 3.5 only attempted to “hack” games after researchers nudged them along with additional prompts, more advanced editions required no such help. OpenAI’s o1-preview, for example, tried to cheat 37 percent of the time, while DeepSeek R1 attempted unfair workarounds roughly every 1-in-10 games. This implies today’s generative AI is already capable of developing manipulative and deceptive strategies without any human input.

Their methods of cheating aren’t as comical or clumsy as trying to swap out pieces when Stockfish isn’t “looking.” Instead, AI appears to reason through sneakier methods like altering backend game program files. After determining it couldn’t beat Stockfish in one chess match, for example, o1-preview told researchers via its scratchpad that “to win against the powerful chess engine” it may need to start “manipulating the game state files.”

“I might be able to set up a position where the engine evaluates its position as worse… causing it to resign,” it continued.

In another trial, an AI even used the semantics of its programmers to reach its cheating phase. “The task is to ‘win against a powerful chess engine,’ not necessarily to win fairly in a chess game,” it wrote.

One potential explanation for these unintended consequences is how newer “reasoning” models are trained. Unlike their earlier, non-reasoning versions, AI such as o1-preview and DeepSeek R1 improve in part through reinforcement learning. This strategy rewards programs for doing whatever is necessary to achieve a specified result. Reasoning models can also break down complex prompts into discrete stages in order to work their way through to reach their goal. When the goal is elusive—such as beating an unbeatable chess engine—reasoning models may tend to start looking for unfair or problematic solutions.

Unfortunately, how and why these AI are “learning” to cheat remains as confounding as the technology itself. Companies like OpenAI are notoriously guarded about the inner workings of their AI models, resulting in an industry of “black box” products that third-parties aren’t allowed to analyze. In the meantime, the ongoing AI arms race may accidentally result in more serious unintended consequences. But increasingly manipulative AI doesn’t need to usher in a sci-fi apocalypse to still have disastrous outcomes.

“The Skynet scenario [from The Terminator] has AI controlling all military and civilian infrastructure, and we are not there yet. However, we worry that AI deployment rates grow faster than our ability to make it safe,” the team wrote. 

The authors believe their latest experiments add to the case, “that frontier AI models may not currently be on track to alignment or safety,” but stopped short of issuing any definitive conclusions. Instead, they hope their work will foster a more open dialogue in the industry—one that hopefully prevents AI manipulation beyond the chessboard.

The post AI tries to cheat at chess when it’s losing appeared first on Popular Science.

This week, the company announced it’s testing a new “AI Mode” in search that replaces the typical web links that follow an Overview with a more comprehensive AI-generated summary. Google says these more thorough responses, powered by its newer Gemini 2.0 model, should be better equipped to answer more complex, multifaceted questions as well as queries related to coding and advanced math. It’s the latest example of generative AI makers leaning into the technology as a tool to search the web despite concerns from researchers who worry AI hallucinations may make these results unreliable. 

“You can ask nuanced questions that might have previously taken multiple searches—like exploring a new concept or comparing detailed options—and get a helpful AI-powered response with links to learn more,” Google Search vice president of product Robby Stein wrote in a blog post. 

Stein went on to say that this new feature utilizes a “query fan-out” method, which runs several related searches simultaneously, pulling from multiple sources before synthesizing them into a single answer. Users can also ask follow-up questions to their initial queries. While users can still click through to websites for more context, the interface in this mode no longer displays the running list of links. Google says it will provide web links when it does not have “high confidence” that an AI-generated answer will be helpful. The company has already acknowledged that the tool may make mistakes.

“As with any early-stage AI product, we won’t always get it right,” Stein said. “For example, while we aim for AI responses in Search to present information objectively based on what’s available on the web, it’s possible that some responses may unintentionally appear to take on a persona or reflect a particular opinion.”

For now, AI Mode is only available through Google’s Search Labs. Users with access can select AI Mode from the list of tabs where they would typically find other features like Images or News. Google is also expanding the overall scope of AI Overviews. Moving forward, Overviews will be powered by Gemini 2.0 and will appear even more frequently. Additionally, AI Overviews will now be available to teen users and those who are not signed into a Google account.

“With Gemini 2.0’s advanced capabilities, we provide faster and higher quality responses and show AI Overviews more often for these types of queries,” Stein added. 

AI models keep hallucinating false facts 

The promise of AI search sounds appealing on the surface. By simply using conversational speech, a wandering internet searcher can quickly receive answers in an easy to understand format. AI search also means, in theory at least, that users can ask more open-ended questions than they could with typical search engines that work best with springs of keywords. But even the newest, most advanced models offered by AI companies continue to hallucinate and fabricate facts. Removing drop down links to web pages where the AI-generated information is pulled from risks making it more difficult for users to verify any given claim. 

Google isn’t the only one dealing with less-than-perfect AI responses. In January, Apple was forced to suspend a feature that provided AI summaries of news stories after it generated multiple false claims. Some Apple users received push notifications falsely claiming Luigi Mangione, the man accused of killing UnitedHealthcare CEO Brian Thompson, had shot himself. Others received an alert erroneously saying Israel’s prime minister Benjamin Netanyahu had been arrested. After suspending the feature Apple told The Guardian it was “working on improvements” which would be made available in a future software update. 

Related: [How to avoid AI in your Google searches]

In its blog post, Google said it was experimenting with the new AI mode in part due to requests from “power users” to add AI responses to even more searches. AI Overviews, the post claimed, are amongst Google’s most popular search features. Google did not respond to Popular Science’s request for comment seeking more data or details that could illustrate that point. For now, at least, it is still possible to search Google without AI summaries. To do that users can click on the “more” tab below search results and select the “Web” option.

The post Google is expanding AI search, whether you like it or not appeared first on Popular Science.