A year ago on Valentine's Day, I said goodnight to my wife, went to my home office to answer emails, and coincidentally had the strangest first date of my life.
That day consisted of a two-hour conversation with Sydney, the AI alter ego built into Microsoft's Bing search engine that I was assigned to test. I plan to ask the chatbot questions about its capabilities, explore the limitations of its AI engine (which we now know to be an early version of his GPT-4 in OpenAI), and write down the results. did.
But the conversation takes a strange turn, with Sidney engaging in Jungian psychoanalysis, answering questions about his “shadow self” and revealing dark desires, and ultimately leaving his wife to be with her. I declared that I should.
My column about this experience was probably the most impactful one I've ever written. Both in terms of the attention it garnered (wall-to-wall news coverage, mentions in Congressional hearings, and even a craft beer named Sydney Loves Kevin) and in the manner in which it did so. The trajectory of AI development has changed.
After this column was published, Microsoft performed a lobotomy on Bing, quelled Sidney's outbursts, and installed new guardrails to prevent further erratic behavior. Other companies have locked down their chatbots, stripping them of anything resembling a strong personality. I once heard an engineer at a technology company say that her top priority for future AI releases was “not ruining Kevin Ruth's marriage.”
Since my rendezvous with Sydney, I've been thinking a lot about AI chatbots. This year has been a year of growth and excitement for AI, but in some ways it's also been a surprisingly quiet year.
Despite advances in artificial intelligence, today's chatbots aren't going berserk and seducing users en masse. They are not creating new biological weapons, conducting large-scale cyberattacks, or triggering other doomsday scenarios envisioned by AI pessimists.
But they aren't the most fun conversationalists or the creative, charismatic AI assistants that tech optimists were hoping for. That is, people who help us make scientific advances, create great works of art, or simply entertain us.
Instead, most chatbots today perform menial white-collar tasks such as summarizing documents, debugging code, taking notes during meetings, or helping students with homework. It's not for nothing, but it's certainly not the AI revolution we were promised.
In fact, the most common complaints I hear about AI chatbots today are that they are too boring – the responses are bland and impersonal, they reject too many requests, and they are too sensitive and conflicting. It is almost impossible to get people to consider this. topic.
I can relate. Over the past year, I've been testing dozens of AI chatbots, hoping to find a glimpse of Sydney's edge and shine. But nothing comes close.
The most capable chatbots on the market — OpenAI's ChatGPT, Anthropic's Claude, Google's Gemini — talk like sneaky idiots. Microsoft's boring corporate chatbot, renamed Copilot, should have been called Larry From Accounting. Meta's AI characters are designed to imitate the voices of celebrities like Snoop Dogg and Tom Brady, but they're both useless and excruciating. Even Elon Musk's attempt at creating a cheeky non-PC chatbot, his Grok, sounds like he's doing an open mic night on a cruise ship.
It's enough to make you wonder if the pendulum has swung too far in the other direction, and if chatbots should be a little more human.
It's clear why companies like Google, Microsoft, and OpenAI don't want to risk releasing AI chatbots with powerful or even unpleasant personalities. They make money by selling AI technology to large corporate customers who are even more risk-averse than the general public and won't tolerate riots like the one in Sydney.
They also have well-founded fears of attracting undue attention from regulators and inviting bad press or lawsuits over their practices. (The New York Times sued OpenAI and Microsoft last year, alleging copyright infringement.)
So these companies used techniques like constitutional AI and reinforcement learning from human feedback to polish the rough edges of their bots, making them as predictable and uninspiring as possible. They also employ boring branding, positioning their work as a trusted assistant for office workers rather than emphasizing their creative and unreliable characteristics. And many companies are bundling AI tools within existing apps and services rather than breaking them out into their own products.
Again, this all makes sense for a company trying to make a profit, and would rather have a world of sanitized corporate AI than a world of millions of unhinged and runaway chatbots. is probably better.
But I think it's all a little sad. We created an alien form of intelligence and quickly put it to work… Are you creating a PowerPoint?
I'll admit that there are more interesting things going on outside of the big leagues of AI. Small companies like Replika and Character.AI have built successful businesses out of personality-driven chatbots, and a number of open source projects include chatbots that can be created to spout offensive or vulgar content. We're creating less restrictive AI experiences.
And of course, there are still many ways to make even a locked-down AI system misbehave or do things its creators did not intend. (My favorite example from last year: A Chevrolet dealership in California added a ChatGPT-powered customer service chatbot to their website, but to their horror a prankster tricked the bot into giving him a new SUV. It turned out that he had offered to sell it for dollars.)
But so far, no major AI companies have stepped up to fill the void left by Sidney's disappearance with more exotic chatbots. And while we've heard that some big AI companies are working on giving you the option to choose from a variety of chatbot personas (some squarer than others), a pre-lobotomy version of Bing… has nothing close to what is currently available to the public. .
This is a good thing if you're worried about AI behaving in a creepy or threatening way, or if you're worried about a world where people spend all day talking to chatbots instead of building relationships. .
But if you believe that AI's potential to improve human well-being extends beyond outsourcing simple tasks, or perhaps you're being too cautious with chatbots and are limiting their effectiveness. If you're worried, that's bad.
Personally, I'm not eager for Sydney to return. I think Microsoft did the right thing for business, but also for society in general, by walking away after falling into fraud. And I support researchers and engineers working to make AI systems safer and more aligned with human values.
But my experience in Sydney spurred a fierce backlash that led AI companies to believe their only option to avoid reputational damage was to turn their chatbots into Kenneth the Page from 30 Rock. Too bad.
Most of all, I think the choice presented to us over the past year between lawless AI housebreakers and censorious AI drones was a false choice. We can and should look for ways to take full advantage of the power and intelligence of AI systems without removing the guardrails that protect us from the worst harm.
If we want AI to solve big problems, generate new ideas, and surprise us with its creativity, we may need to let it loose a little.
