If you’re new here, this is the latest edition of the Build Mode newsletter. Subscribe here to get the top insights in your inbox every week.
The Big Idea: Will 2024 see the first AI employee?
Are you ready to let ChatGPT use your computer for you?
According to The Information, a new AI agent will have the power to take over your device, use the cursor to click around, type text into forms, and do whatever you ask it to—like take an unwieldy document, turn it into a nice crisp spreadsheet, and then analyze the data.
In addition to this computer control agent, there’s also a web browsing agent which can book flights, gather data about companies, and other complex tasks.
Creepy? Maybe a little. But the possibilities for this are endless. Everyone can have a personal assistant—or 100—working around the clock. As the technology improves, you can ask for increasingly sophisticated tasks. Which is how you end up with Sam Altman and his buddies taking bets on when the first 1-person billion dollar company will arrive.
ChatGPT is already introducing new memory features such that the chatbot will remember you and your preferences from previous chats. This is huge. Imagine having an assistant who remembers better than you do what you’ve been working on.
Plus there’s evidence that LLMs can display better strategic decision making than humans. In a recent paper, ChatGPT’s decisions were largely rational in risk, time, social, and food choices and demonstrated higher rationality scores than those of humans. This means that an agent could really be fully autonomous soon, not just automating tasks, but making strategic decisions all on its own.
Vivan Amin, an AI leader at L3Harris Technologies and a member of our Editorial Board, predicts that we’ll see the first AI agents operating as fully autonomous workers by the end of the year.
“Once it starts making these multimodal decisions based on all the inputs and the data it has—that's when the agents go from agents to employees.”
Where will this happen first?
“One sector that's already getting disrupted is customer service. The other one is finance. Robo-advisors, right? We already have those. Because I don’t need to pay someone a massive amount of my portfolio to tell me what my tax benefits should be,” Amin said. “But you’ll still need humans to talk to humans when it comes to large amounts of money.”
Companies have directly attributed more than 4,600 layoffs to AI since May, according to research from the outplacement firm Challenger, Gray & Christmas. But there’s a difference between needing fewer workers because of AI and employing AI workers full time.
Which, as the following chart shows, is slightly lower than anticipated.
Missed last week’s issue of Build Mode? Read it here.
CHART OF THE WEEK
AI is still too expensive to replace humans in most jobs
AI is still too expensive to replace humans in most jobs, a recent working paper from MIT and the IBM Watson AI Lab found.
As Michael Spencer and Tobias Mark Jensen explain in the AI Supremacy Substack: “Previous studies on AI’s labor market impacts have assumed that any work task an AI system is capable of performing is exposed to automation without considering the technical feasibility or the economic viability of automating such tasks. This is quite an important detail. Just because a work task can be automated does not mean that it is desirable or feasible to do so from an economic and/or technical perspective.”
So the researchers distilled this complex problem down to two questions:
- Exposure: Is it possible to build an AI model to automate this task?
- Economically-attractive: Would it be more attractive to use an AI system for this task than to have human workers continue to do it?
More specifically, they mainly looked at the costs of fine-tuning and deploying a computer vision system to perform a task, including fixed costs, performance-dependent costs, and scale-dependent costs in a given time span vs. the marginal cost of compensation per worker.
They found that only 23% of visual-based tasks would be cost-effective for firms to automate. Put that in contrast to McKinsey’s prediction that activities accounting for 30% of hours worked in the US could be automated by 2030.
Another factor: LLMs will become cheaper to run in the coming years. But even with a 50% annual decrease in costs, the researchers estimate that it will take until 2026 before half of the vision tasks have an economic advantage.
AROUND THE WATERCOOLER
ChatGPT passes the GARP test
The classic test: Would you take one marshmallow now? Or two in five minutes?
ChatGPT knows the right answer. And there’s a new paper showing how LLMs might be better at decision-making than humans.
The Generalized Axiom of Revealed Preference or GARP is a measure of how logically someone balances trade-offs, like whether to spend money today vs. save for a rainy day.
The study found that 81-95% of ChatGPT-3.5’s decisions in domains like risk, time, social, and food respect GARP, indicating that ChatGPT shows some serious rational decision-making chops, often outperforming us humans.
ChatGPT also tended to be less biased than humans when contemplating two options, bypassing the emotional baggage, past experiences, and shortcuts that usually sway human judgment.
DEEP DIVE
4 CFOs on the path to ROI for AI initiatives
HungryRoot, the New York-based grocery subscription service, has devised a tasty use for artificial intelligence. Customers feed the company information about their households' dietary needs. In return, AI-powered tools help HungryRoot feed its subscribers by devising personalized meal plans and grocery orders that the company ships out every week.
The AI investment is even palatable to the company's finance department: data shows that more personalized orders lead customers to buy more and cancel subscriptions less, says Wajeeha Ahmed, HungryRoot's CFO. AI can tailor customers' orders by synthesizing their personal requirements (a gluten allergy) with external signals (a blizzard approaching their zip code).
Cozy chicken stew with rice noodles, anyone?
"There's a very direct business result," Ahmed told an audience at an A.Team salon on The ROI of AI, on January 31 in New York City.
She joined three other seasoned CFOs—Vadicel Joy Abboud of Ceremonia, Chris Asmis of Farmer's Dog, and Varun Athi, formerly the CFO of Luma, Aventri, and Butterfly—to sift through the risks and rewards attached to AI investment.
In 2024, Generative AI is widely viewed as a magic ingredient. Sprinkled into existing software – or served separately via subscription products like ChatGPT – GenAI-powered tools can handle all manner of specialized tasks, potentially boosting productivity, eliminating repetitive work, improving customer service, and (presumably) fattening bottom lines.
EVENTS
Staying Human in the Age of AI
As many tech leaders roll out generative AI initiatives, they're facing a big problem: employees are freaked out.
While the vast majority of execs are optimistic about AI, two-thirds of employees are scared. This creates a serious issue: You can have the best tech in the world, but if your employees aren't bought in, digital transformation fails.
At our next Generative AI Salon on Wednesday, February 28th, we’ll explore how to get AI right. Hosted by four luminaries from the A.Team’s CxO Network, we’ll reveal how to drive AI transformation from four perspectives:
- Organizational: How to leverage AI to empower our workforce and build the organization of the future, with Wagner Denuzzo.
- Teams: How AI can enable our teams to do their best work, with AJ Thomas.
- Culture: How to create a culture of Collective Intelligence, with Angelique Bellmer Krembs.
- Individual: How to gain clarity into your own personal ambitions around AI adoption, with Dr. Eric Solomon.
After the panel, we'll shift gears into an interactive session where you'll have the opportunity to craft an AI plan that you can put into action.
DISCOVERY ZONE
A study predicted that 45% of men would use AI to write their Valentine's cards this year — we can only hope they used Provenance, an AI tool that takes your personal experiences and translates them into the right words for the occasion.
DEEP DIVES FROM THE ARCHIVES
- What the Science of Creativity Tells Us About Writing with Generative AI
- 3 Things Generative AI Can Never Do Better Than Humans
MEME OF THE WEEK