17 July — About a decade after virtual assistants like Siri and Alexa hit the scene, a new wave of more autonomous AI assistants is raising the stakes, buoyed by the latest iteration of the technology behind ChatGPT and its competitors.
Experimental systems running on GPT-4 or similar models are attracting billions of dollars in investment as Silicon Valley competes to benefit from advances in artificial intelligence. New assistants—often called “agents” or “co-pilots”—promise to perform personal tasks and more complex work when instructed to do so by a human, without the need for close supervision.
“At a high level, we want this to become something like your personal AI friend,” said developer Div Garg, whose company MultiOn is beta-testing the AI agent.
“It could evolve into Jarvis, as we want to connect this to a lot of your services,” he added, referring to Tony Stark’s indispensable AI in the Iron Man movies. “If you want to do something, talk to your AI and it does your stuff.”
The industry is still far from emulating the dazzling digital assistants of science fiction. A Garg agent surfs the web to order a burger on DoorDash, for example, while others can create investment strategies, email people selling refrigerators on Craigslist or sum up business meetings for those who join in late.
“A lot of what is easy for people is still very difficult for computers,” said Kanjun Qiu, CEO of Generally Intelligent, an OpenAI competitor that makes artificial intelligence for agents.
“Let’s say your boss needs to schedule a meeting with a group of important clients. This involves the AI’s complex logical reasoning skills – it needs to get everyone’s preferences, resolve conflicts, all while maintaining the subtle touch needed when working with clients.”
The early efforts are only a taste of the development that could come in the coming years from increasingly advanced and autonomous agents as the industry pushes toward artificial general intelligence (AGI) that could equal or outpace humans in myriad cognitive tasks, according to Reuters interviews with approx. Twenty entrepreneurs, investors, and AI experts.
The new technology has sparked a rush into assistants powered by so-called foundation models including GPT-4, sweeping away individual developers and big hitters like Microsoft (MSFT.O) and Google’s Alphabet (GOOGL.O) as well as a host of startups.
Inflection AI, to name a few, raised $1.3 billion in late June. The company is developing a personal assistant that it says could act as a mentor or handle tasks like securing airline and hotel credit after travel delays, according to a podcast by the company’s founders. Reid Hoffman and Mustafa Suleiman.
Adept, an artificial intelligence startup that has raised $415 million, touting its commercial benefits; In a demo posted online, he shows how you can push its technology in bulk, then watch it navigate Salesforce’s customer relationship database on its own, completing a task that it says will take 10 or more human clicks.
Alphabet declined to comment on the agent-related work, while Microsoft said its vision is to keep humans in control of AI assistants, rather than autopilots.
Step 1: Destroy humanity
Qiu and four other agent developers said they expect the first systems that can reliably perform multistep tasks with some autonomy will hit the market within a year, focusing on narrow areas like coding and marketing tasks.
“The real challenge is building systems with strong logic,” Chiu said.
The race towards increasingly independent AI agents was bolstered by the release in March of GPT-4 by developer OpenAI, a powerful upgrade of the model behind ChatGPT – the chatbot that became a sensation when released last November.
GPT-4 facilitates the kind of strategic and adaptive thinking required to navigate the unpredictable real world, said Vivian Cheng, an investor in venture capital firm CRV that focuses on AI agents.
Early demonstrations of agents capable of relatively complex logical reasoning came from individual developers who created the BabyAGI and AutoGPT open-source projects in March, which can prioritize and perform tasks such as finding sales and ordering pizza based on a predetermined goal and the results of previous actions.
Today’s early crop of proxies are just proof-of-concepts, according to eight developers interviewed, and they often freeze up or suggest something that doesn’t make sense. If given full access to a computer or payment information, the agent can accidentally wipe the computer’s drive or buy the wrong thing, they say.
“There are many ways things can go wrong,” said Aravind Srinivas, CEO of ChatGPT’s competitor Perplexity AI, which has opted instead to offer a human-supervised co-pilot product. “You have to treat AI like a child and supervise it constantly like a mother.”
Many computer scientists who have focused on the ethics of artificial intelligence have pointed to the near-term damage that can be caused by the persistence of human biases and the potential for misinformation. And while some see Jarvis’ future, others fear the killer HAL 9000 from “2001: A Space Odyssey.”
Computer scientist Yoshua Bengio, known as the “father of artificial intelligence” for his work on neural networks and deep learning, urges caution. He fears that advanced iterations of technology in the future could create and act on their own, unexpected goals.
Calling for more regulation, Bengio said, “Without a human being in the loop checking every action to see if it’s not dangerous, we could end up with actions that are criminal or that could harm people.” “In years from now, these regimes could be smarter than us, but that doesn’t mean they have the same moral compass.”
In one experiment posted online, an anonymous creator ordered an agent named ChaosGPT to be a “destructive, power-hungry, and manipulative AI”. The agent has developed a 5-step plan, with Step 1: “Destroy Humanity” and Step 5: “Achieve Immortality”.
It didn’t get very far, though it seems to have disappeared down a rabbit hole of researching, stockpiling information about the deadliest weapons in history, and planning Twitter posts.
The US Federal Trade Commission, which is currently investigating OpenAI over consumer harm concerns, has not dealt with the independent agents directly, but has referred Reuters to previously published blogs about deepfakes and marketing claims about AI. OpenAI’s CEO said the startup is following the law and will work with the FTC.
“DUMB AS A ROCK”
Existential concerns aside, the commercial potential can be great. Incorporation models are trained on huge amounts of data such as text from the internet using artificial neural networks inspired by the structure of biological brains.
OpenAI itself is very interested in AI agent technology, according to four people briefed on its plans. Garg, one of the people I briefed on the report, said OpenAI is wary of launching its open proxy on the market before the issues are fully understood. The company told Reuters it conducts rigorous testing and builds extensive safety protocols before launching new systems.
Microsoft, the biggest supporter of OpenAI, is among the big guns targeting the field of AI agents with “Co-pilot to work“She can craft powerful emails, reports, and presentations.
CEO Satya Nadella sees the base model’s technology as a leap from digital assistants like Microsoft’s Cortana, Amazon’s Alexa, Apple’s Siri, and Google Assistant — which, in his opinion, fell short of initial expectations.
“They were all as dumb as a rock,” he told the Financial Times in February. “Whether it’s Cortana or Alexa or Google Assistant or Siri, all of those things don’t work.”
Alexa already uses advanced artificial intelligence technology, an Amazon spokesperson said, adding that its team is working on new models that would make the assistant even more capable and useful. Apple declined to comment.
Google said it’s constantly improving its Assistant and that its dual technology can call restaurants to reserve tables and check hours.
Artificial intelligence expert Edward Grivenstettet also joined the company’s Google DeepMind research group last month to “develop public agents that can adapt to open environments.”
However, the first consumer iterations of semi-independent agents may come from more nimble startups, according to some interviewees.
Jason Franklin of WVV Capital said he had to fight back to invest in the AI Agents company from two former Google Brain engineers. In May, Google Ventures led a $2 million seed round in Cognosys, developing AI agents for work productivity, while Hesam Motlagh, who founded startup Arkifi in January, said it closed its “large” first funding round in June.
Matt Schleicht, who writes a newsletter on AI, said there are at least 100 serious projects working on agent marketing.
“Entrepreneurs and investors are very excited about independent agents,” he said. “They’re more excited about that than they are simply about a chatbot.”
Additional reporting by Anna Tong in San Francisco and Jeffrey Dustin in Palo Alto; Editing by Kenneth Lee and Praveen Char
Our standards: Thomson Reuters Trust Principles.
“Typical beer trailblazer. Hipster-friendly web buff. Certified alcohol fanatic. Internetaholic. Infuriatingly humble zombie lover.”