The AI Daily Brief: Artificial Intelligence News and Analysis - How Much AI Do Workers Actually Want?

Starting point is 00:00:00 Today on the AI Daily Brief, how much AI do workers actually want? And before then, in the headlines, Anthropic goes vertical with Claude for financial services. AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. All right, friends, quick announcements today. First of all, thank you to today's sponsors, blitzie, superintelligent, and agency.org. And to get an ad-free version of the show, go to patreon.com.com slash AI Daily Brief. Speaking of brief, let's keep the announcements brief today and dive right in. Welcome back to the AI Daily Brief Headlines edition,

Starting point is 00:00:34 all the daily AI news you need in around five minutes. Boy, is it a tough time to be a startup? On the one hand, AI feels so full of possibility. We are in a transition point between two totally different paradigms, and that's always the best most opportune time to build something big and new. However, whereas in the past, you could kind of wave off the 800-pound gorillas in the room of the former incumbents as two lumbering and slow-moving.

Starting point is 00:00:59 In the case of AI, startups are dealing not only, with the Mag 7 type competitors, but with the foundation model companies specifically OpenAI and Anthropic, increasingly seeming to recognize that it really matters to own the relationship with their customers through the application layer. The latest example of that is Anthropic turning Claude into a financial analyst. Now, this is an example of Anthropic following a use case that they're seeing. Many people are starting to turn to LLMs to provide market analysis, or for an easy way to run calculations like price to earnings that enable them to evaluate stocks. Proplexity has already leaned into this trend by making financial analysts a core part of recent releases,

Starting point is 00:01:37 but Anthropic has taken the idea step further. Rather than a consumer-facing product, Claude for Financial Services is a specifically designed version of their enterprise subscription. It includes built-in MCP support for major data providers and financial industry platforms, making it their first vertical product, and definitely creating a template for them to do this for other industries and areas. Anthropics suggest that the model can speed up due diligence and research, generate financial models with full audit trails, and help with portfolio management and benchmarking. It's also clearly a test for Anthropic to see how big the market is for industry-specific platforms. Said Jonathan Pelosi, the head of industry for financial services at Anthropic, the company's models were already

Starting point is 00:02:15 particularly well-suited for financial workloads, and we've been tailoring them to get better and better. Interestingly, he continued, unlike some competitors in the space who built a consumer app that became a sensation, or they built these new video generators or meme generators, that's just never our focus. We are enterprise first, so our models perform best in class against complicated enterprise workloads, which means complicated qualitative analysis and complex data extraction for the financial services industry at scale. Now, I don't know how much this is Pelosi just speaking on his own turn, or whether this is actually the new narrative coming out of Anthropic, but to the extent that it's the latter,

Starting point is 00:02:49 that would be pretty interesting to me. It's not totally new for Anthropic to be leaning into their enterprise use case, but putting it so bluntly and clearly would be a bit of a shift that suggests that that's where they think their opportunity lies. Ultimately, the fingerprints of verticalized applications from the foundation model companies is all over this. Said Pelosi, think of it as an out-of-the-box solution that's easily configurable for the Bridgewater's or the Norwegian sovereign wealth funds of the world, versus the alternative where they cobble this thing together on their own. Next up, Google has flipped the AI switch on Google Discover, accelerating their push towards generated content. Google Discover is the lightweight news feed inside the company's mobile products, where it previously,

Starting point is 00:03:29 served a new headline and a link to the original source, Google will now present an AI summary of major news items. Users will see the logos of multiple news sources in the top left-hand corner of the card if they want to read more, but this is still very different than the previous model. Now, the change is still early in its U.S. rollout, but Google confirmed that this is a proper launch and not a test. Earlier this week, the Economist picked up the story of Google's AI features driving web traffic down. They cited statistics from similar web that showed global web traffic is down by 15% year-on-year, and zero-click news searches had grown from 56 to 69%. The headline declared, AI is killing the web. Can anything save it?

Starting point is 00:04:06 Next up, an update on a story that we've been following. Former OpenAI CTO Mira Muradi's thinking machines labs has officially closed their seed round at a $12 billion valuation. Mira has famously been raising money without a product or detailed public roadmap, instead selling investors on talent, presenting a team filled with researchers drawn from the leading AI labs. The round raised $2 billion and appeared to get a valuation bump from the rumors of a $10 billion price tag last month. More likely, the reports had the number right, but were just confused about whether that $10 billion valuation was pre-money or post-money. And recent Horowitz led the investment with participation from Nvidia, Excel, Service Now, Cisco, AMD, and Jane Street. One of the

Starting point is 00:04:45 largest seed rounds in Silicon Valley history, this demonstrates that AI talent is clearly the hottest commodity in the current venture environment. Investors were even willing to forego board seats in Grant Maradi a controlling vote in order to close the deal. Upon the announcement, Mora tweeted, Thinking Machines Lab exists to empower humanity through advancing collaborative general intelligence. We're building multimodal AI that works with how you naturally interact with the world, through conversation, through sight, through the messy ways we collaborate. We're excited that in the next couple of months we'll be able to share our first product, which will include a significant open source component and be useful for researchers and startups developing custom

Starting point is 00:05:18 models. Soon we'll also share our best science to help the research community better understand frontier AI systems. So ultimately, we still don't really know what they're doing. But it is notable that in the time since that deal has come together, meta has gone on its full-on assault on the talent space. It appears that two more extremely senior researchers at OpenAI are now going over to Meta as well. So this competition gets nothing but more and more fierce. For now that is going to do it for today's AI Daily Brief Headlines edition.

Starting point is 00:05:46 Next up, the main episode. This episode is brought to you by Blitzy. If you're a technology leader, here's something that probably sounds familiar. Your organization's competitive edge is buried in legacy code that desperately needs modernization, but the resources required feel out of reach. That was the case for a global investment analysis firm. They needed to migrate 70,000 lines of complex Matlab financial algorithms to Python, algorithms that drive investment decisions for trillions in assets. Their estimate, months of high-cost specialized engineering work. Instead, they partnered with Blitzy.

Starting point is 00:06:19 Blitzie's autonomous AI preserved mathematical precision and generated over 80% of the new codebase, completing the migration with just five days of engineering time. They cut the timeline by 95% and saved 880 engineering hours. If your organization is facing similar modernization challenges, visit blitzie.com to schedule a consultation and discover how AI power development can transform your technical capabilities. Today's episode is brought to you by superintelligence, specifically agent readiness audits.

Starting point is 00:06:47 Everyone is trying to figure out what agent use cases are going to be most impactful for their business, and the agent readiness audit is the fastest and best way to do that. We use voice agents to interview your list. leadership and team, and process all of that information to provide an agent readiness score, a set of insights around that score, and a set of highly actionable recommendations on both organizational gaps and high-value agent use cases that you should pursue. Once you've figured out the right use cases, you can use our marketplace to find the right vendors and partners, and what it all adds up to is a faster, better agent strategy.

Starting point is 00:07:20 Check it out at besupor.a.i or email agents at besupor.a.i to learn more. Today's episode is brought to you by agency, an open source collective for interagent collaboration. Agents are, of course, the most important theme of the moment right now, not only on this show, but I think for businesses everywhere. And part of that is the expanded scope of what agents are starting to be able to do. While single agents can handle specific tasks, the real power comes when specialized agents collaborate to solve complex problems. However, right now there is no standardized infrastructure for these agents.

Starting point is 00:07:55 to discover, communicate with, and work alongside one another. That's where agency, spelled A-G-N-T-C-Y, comes in. Agency is an open-source collective building the Internet of Agents, a global collaboration layer where AI agents can work together. It will connect systems across vendors and frameworks solving the biggest problems of discovery, interoperability, and scalability for enterprises. With contributors like Cisco, Crew AI, Langchain, and MongoDB, agency is breaking down silos and building the future of Internet.

Starting point is 00:08:25 interoperable AI. Shape the future of enterprise innovation. Visit agency.org to explore use cases now. That's agn tcY.org. Welcome back to the AI Daily Brief. One of the obvious tensions to be resolved when it comes to artificial intelligence and work is the potential for misalignment

Starting point is 00:08:44 between what management wants out of AI and what workers want out of AI. And the challenge to some extent runs along the spectrum of automation, where there might be broad alignment between management and workers around automating menial tasks? When you get to the point where you're actually automating away people's jobs, obviously there's going to be disagreement. Now, a new study from Stanford has taken a novel approach to AI disruption at work

Starting point is 00:09:06 by just straight up asking workers what they actually want. And this comes at an opportune time. Even though we are just two and a half years post-chat GPT, we are already starting to see the effects of AI disruption. We've seen multi-decade lows in hiring for college graduates, waves of layoffs at big tech firms, although, as we always discuss, there are larger macro forces at work there, recalibration post-COVID, and other countervailing forces, but still it's very clear that they're positioning for a world in which you need less people to do the same amount of work.

Starting point is 00:09:37 And as much as worker displacement captures the headlines, the reality is, for the vast majority of workers, the balance of this decade, and frankly into the next, is going to be about learning to work alongside AI. And given that, Stanford researchers from the Stanford University human-centered artificial intelligence lab, HAI, interviewed more than 15,000 workers across more than 100 occupations to uncover which tasks they thought AI could handle and where AI replacement would actually be harmful. Now, alongside the survey, researchers also talked to around 50 AI experts to better understand the current capabilities of AI and how they line up with the desires of workers. This resulted in a huge range of insights in where the opportunities lie for AI

Starting point is 00:10:15 at work and some of the pitfalls for startups looking to address this space. Framing the research, co-author De Yang said, as the workforce evolves, understanding and bridging the gap between worker expectations and the realities of AI capabilities will be crucial for organizations striving for successful integration. This report offers a timely and structured baseline of where we are right now. Now, the best way to understand the findings is this chart, which if you are listening to this not watching it, I suggest you go to Spotify or YouTube just to check it out quickly. It's one of the most information dense charts of any AI research paper I've come across. It maps each use case identified in the survey with its current feasibility according to the AI experts.

Starting point is 00:10:54 The chart is split into four quadrants. The automation greenlight zone is where feasibility meets a desire for automation, i.e. where we can automate things and where people want things to be automated. In the R&D Opportunity Zone, there are tasks where there's a strong desire for automation, but the tech is still lagging behind a little bit. The low priority zone contains tasks where workers don't want automation and the work is difficult for AI complete. And finally, there is the Automation Red Light Zone, where tasks can feasibly be done by AI, but workers do not want to see a computer take over for various reasons. Now, let's start in the Automation Red Light Zone.

Starting point is 00:11:30 Some of the examples of tasks they give are researching hardware and software for network support specialists, contacting vendors regarding material availability for logistics analysts, and preparing meeting agendas for court workers. These are all extremely sensitive tasks where hallucinations and other errors are simply not acceptable. Failure carries a high cost, so workers want to apply their own skill to the task. Now, one thing to note is how important the environment or context for a task is. It's not just about whether AI can do the thing, it's about how relatively important doing that thing right is in the context of the work.

Starting point is 00:12:06 For example, in a lower stakes environment, current AI is perfectly capable of carrying out many of these types of tasks. You wouldn't think twice, for example, about using AI to research hardware and software while building a gaming PC, or preparing an agenda for a run-of-the-mill weekly corporate meeting. But if a hallucination during logistics work means your multi-billion dollar project doesn't have enough steel, you're probably going to want a human on the job. Now, one really interesting observation about this came from Angel Anuham, the CEO of Avella's health. Angel noted that 41% of YC startups are currently building in this red zone.

Starting point is 00:12:39 He suggested that this signals a disconnect between founders and customers. They might see a certain workflow, think that AI-Cube, can easily carry it out, but fail to understand why workers in that specific context might not want AI anywhere near that particular job. Now, Angel points out that the obvious reason for this is that this is just where the average youth and inexperience of a young startup founder runs into the reality of the real world. If you don't actually have deep domain knowledge, you might not realize that the task in that particular industrial context is more significant than that task in other types of contexts. Now, ultimately, where YC is building is second,

Starting point is 00:13:15 I think to the larger part of the study, but it is an interesting outcome to be able to better map where different startup opportunities lie as well. Now, one other thing that comes up from the automation red light zone is that we run into challenges the more binary our thinking is about automation. What it feels to me like workers are responding to in many of these cases is the idea of being automated out entirely. However, there are clearly opportunities for human augmentation with AI and agents even in these examples, i.e., you might not want to fully automate highly specialized hardware research end-to-end, but there are certainly going to be tools and functions that allow the people who are doing that work to move more quickly. As part of the study, the Stanford team

Starting point is 00:13:53 introduced what they call a human agency scale. It runs from H-5 to H-1. On the H-5 end of the spectrum, the human drives task completion. The human takes primary responsibility for task execution with varying levels of AI assistance. In the middle at H-3, there is an equal partnership between human and AI. For HAS H1 and H2, the agent is actually the force-driving task completion. They consider H3 to H5 to be human augmentation where AI enhances human capabilities, and H1 and H2 to be automation where AI replaces human capabilities. Now, there are lots and lots of different versions of this scale. We recently did that episode about the seven different types of AI agents, and you can map this against some of that as well. The point is that thinking in binaries is going to be

Starting point is 00:14:37 problematic, and thinking about automation and augmentation as a spectrum is going to be much more useful in practice. Now, moving back to the other zones, they're far less controversial. For example, the low-priority zone, where things are hard for AI and humans don't want AI involved, one example they give is ticket agents that trace lost, delayed, or misdirected baggage for customers. Luggage handling already has a great deal of automation involved, so if something has already gone wrong, workers understandably don't want AI handling the problem. Basically, by the time that they're getting called in, some version of AI has already screwed up. Another example that's given is art directors who don't want AI presenting the final layouts to clients. This to me is an

Starting point is 00:15:15 invocation of taste, and the idea that art directors want to be able to explain decisions in a way that involves taste and choices which might be unexpected. How many times have you heard of some brand or creative presentation where the creative presents something and initially the client hates it, only to come around later for it to be the thing? Slightly different than art directors, but an example of this that I heard recently was Steven Spielberg talking about John William's score for Jaws. John invited Stephen over to his house all excitedly to share with him his idea for the score, and he played him the now iconic, da-da notes. And Spielberg was like, what the heck is this? Where's the top line? Where's the melody? William stuck to his guns and obviously it became one of the

Starting point is 00:15:54 most iconic scores of all time. Now, moving on to the green zone, it's sort of the opposite of the low priority zone where this is obvious areas of opportunity. This is where AI automation do good work and where people desire it to do good work. The examples they gave are mechanical engineers reading and interpreting reports, tax preparers scheduling appointments, and quality control managers checking routine data, all very obvious in some of the use cases that are already being deployed. Some of the most interesting opportunity is in the aptly named R&D Opportunity Zone.

Starting point is 00:16:23 These are areas where workers really want automation and augmentation, but where the perception is, at least, that AI isn't up to the job yet. The examples given in the study were a computer scientist managing operational budgets, technical writers arranging distribution of materials, and video game designers creating production schedules and managing prototyping goals. Now, the specific examples given are fairly niche, but they also generalize out to some pretty broad areas of workflows that can be duplicated across different industries. Now, the list is far from exhaustive, but if you are digging for startup ideas, it might be worth reading the full study to see if anything clicks.

Starting point is 00:16:55 Zooming back out from the individual tasks, the researchers put forward some broad takeaways from their data. At the top line, the study reconfirmed that workers don't want to be replaced. but they do want AI to lighten their load. 69% of workers said that they welcomed automation that would free them up for higher value tasks. 46% said that they wanted automation to reduce task repetitiveness with the same proportion saying they want automation to improve the quality of their work.

Starting point is 00:17:17 Only 2% of workers said that they wanted full automation that requires no human input, while 35% said that they wanted to see automation with human oversight at key junctures, and 45% said that they wanted a roughly equal partnership between human and AI work on automated tasks. The study also revealed that workers are pretty positive about AI adoption as long as they have a stake in how it happens. Respondents had positive attitudes towards AI augmentation for 46% of tasks overall.

Starting point is 00:17:43 The key concern also wasn't job replacement with just 23% of respondents listing that as their fear. Instead, 45% of respondents said that they had a lack of trust in AI accuracy, capability, or reliability. Basically, workers don't want to be saddled with AI tools that they didn't have a choice in and don't do the job well. The researchers also drew some conclusions about which skills would be valued in the future. They believe that data analysis, currently one of the highest paid skills, is in for a big correction lower as AI takes over. The skills they believe will be more highly valued, on the other hand, were associated with management, including organizing or planning work and training others. To take the broad strokes, they believe that tasks that require a high level of expertise but

Starting point is 00:18:19 are relatively formulaic are going to be devalued, things like process monitoring and industrial settings. By the same token, soft skills that involve effective communication and empathy are going to be much more highly valued. Again, co-author D.E. Yang says, said, an increased emphasis will be placed on skills that require human interaction and coordination. Now, one of the big contentions of the paper was that the opinions of workers really matter in guiding the AI overhaul of the workplace. Ultimately, employees are the ones that go through the workflows day in and day out, so they're much better equipped to see where tedious or repetitive tasks can be automated away. This is something that we have found over and over again when it

Starting point is 00:18:54 comes to the agent readiness audits that we do at super-intelligent. The whole idea of architecting these things to be driven by voice agents is that for the first time, we can actually interview everyone. If you think about how a traditional consulting firm like a McKinsey would have to do the type of discovery that we do, they might use a combination of interviews and surveys. Traditional interviews are great for context but bad for scale because they take so much human time and effort, while surveys are good for scale but bad at context, there's limits to how much information you can get from multiple choice or people typing out their answers. Voice agents that can actually interview everyone across an organization all at the same time, is a totally new capability

Starting point is 00:19:33 unlocked by AI that allows us to get incredibly rich data from the ground up across the entire organization. What we tend to find is that our biggest value is not some blistering insight around agentic use cases. It's the fact that when we suggest a use case, we're adding pull quotes from Marion Accounting, Stephen Marketing, and Jenny in sales that all point to that same reality. In other words, it's not really super intelligent that's making agent recommendations. It's the people who are at the company who are closest to the work. Now, even if the world moves to a model where worker opinions are deeply valued around AI adoption, there are still going to be challenges.

Starting point is 00:20:09 The inescapable reality is that as agents get more capable, there are going to be buckets of tasks that get so fully automated that it will implicate certain people's jobs. My humble suggestion is that the more that management can do now to engage in incorporation and incorporate worker voices from up and down the organization, the better able to navigate those complexities they're going to be as those realities come up. Pretending those things aren't happening is not the answer.

Starting point is 00:20:34 Getting people involved in the remaking of their own organization seems to me to be a much better path. I think studies like this one go a long way to helping us get more granularity and specificity around some of these questions. So great work to the team at Stanford who led this project. And thanks, of course, to you guys for listening or watching. That's going to do it for today's AI Daily Brief. Until next time,

Starting point is 00:20:53 Peace.

The AI Daily Brief: Artificial Intelligence News and Analysis - How Much AI Do Workers Actually Want?

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.