ML good, AI bad, Humans ugly?

Almost every startup now has an obligatory AI component, even when it's not merited. There are various reasons for this. Firstly, it's because the early stage investment community is now disinterested in anything that doesn't have AI involved in some way - if you want to get invested, you gotta have it (apparently). Secondly, it comes from either deliberately or ignorantly misdefining features as AI. Finally, it comes from not understanding where the value in the service proposition comes from - and much capital is being destroyed from all three of these reasons.

Value proposition

I want to start here because it's the most important place to start for any new business. We should be clear what we are trying to do, and why. Mission, vision, etc. If at your core, you have a great idea for disrupting the insurance marketplace or travel planning and brokerage, the question is whether your target market wants efficient service, low prices, or abundant imagination and novelty in what you are doing. For simple, low ticket size insurance products, the answer surely is operational efficiency. Here, use of algorithms integrated with the new LLM-based agents makes a compelling case for wholesale adoption of these new technologies. Algorithms for fraud detection, LLMs for customer service, with escalation to humans for edge cases. Now, for complex, high value coverage, the expectation will be for a much "higher-touch" experience. So the product may seem the same, but the service expectation is wildly differnt.

For travel, one might want simply to buy a ticket from A to B and pricing is everything, so AI adds nothing. Indeed, making changes to a booking may warrant use of an agent, but early experience shows that frankly it's a waste of time, energy, and human input is required regardless. This will no doubt get better over time, but the cost/benefit calculation will be interesting when we see that all AI model providers are losing money at the moment - to become sustainable the unit economics will have to be transparently better than the human touch. The jury is still out on this. When considering travel agency and experiences, the human involvement and creativity to propose new and exciting plans for a future trip - again an LLM will give you a statistically significant answer to your question, but will it tickle your fancy, or excite you?

Bear in mind, these AI agents are based on input material scraped from existing human content and creativity. Yes, you can add "temperature" or randomness to generate creative tone to the answers offered, but the source IP is still stolen. After decades of US and European tech firms complaining about Chinese companies stealing or copying their IP, we have industrial scale lobbying from the current generation of tech titans demanding that they be allowed to steal creative content from fellow humans on a scale not seen before. We have to ask ourselves if this is sustainable, let alone morally acceptable.

The great AI lies

Without doubt, proponents of the implementation of some new tech are getting ahead of their skis because they see the future possibilities, but this does not mean use of AI is always appropriate. To cut the wheat from the chaff, we first must step back and check that we are speaking of the same thing.

AI is as poorly defined today as any tech at the beginning of a hype cycle. Is it merely a difference or statistical engine capable of brute forcing an answer to a human question to the extent that it passes the Turing Test?

Next to that we have ML - where statistical models can be generated against large data sets specific to an individual use-case. Use of ML to provide assistance in the interpretation of X-rays or MRI scans is of significant value. Given the shortage of trained radiologists, this doesn't replace the art of radiology, but perhaps means the job will change from that of being a cog in the diagnostic machine, to one more geared towards research and development (blue sky thinking, which humans are good at), or interventions (which even with robot-assisted surgery, remains the domain of humans).

Investors

I do wonder if we are incapable of learning from history.

The current investor landscape shows no signs of remembering the dot-com bubble. I watch with despair as startups with great technology - worse yet, truly novel science - fail to get funded because they are not in the AI race. Over 90% of VCs (based on a recent virtual roadtrip I participated in) who claim to invest in health or life sciences will only invest if it comes at the "intersection of data and science". This is all well and good, but we now see ourselves in a situation where great tech is not being funded, whilst a flurry of activity is happening where we are spitting out tens of thousands of candidate molecules in pharma, and now having to figure out how to select down to find the needles in the haystacks.

Add to this the datacenter and energy investments being made, AI is proving to be a black hole, sucking in capital and talent from anywhere and everywhere.

From startups threatening to publish 8000 books a year using gen AI, to the announcements of social media bots being driven by AI, it's unclear how much of this investment is socially useful, and what the cost in terms of time, money and opportunity cost will be with capital being deployed on AI only ventures rather than "real businesses".

Over $5tn of capital was lost in the bubble over 20 years ago. Given that NVIDIA alone lost $600bn in market cap in a single day last week thanks to the announcement of more efficient LLM models from DeepSeek, the potential for a bonfire of the vanities is a real and present danger.

ML good, AI bad, Humans ugly?

Where does this leave me? The answer is, frustrated.

The opportunities from ML are real and already clear. The efficiency and improvement in so many spheres from use of modern data science is genuinely exciting and we are just at the beginning. The current frenzy around AI however feels like a buzzy distraction. The idea of communicating in natural language with data, algorithms, and dynamic systems ans services is exciting, however our understanding of where the value truly is appears poor. 24 years ago, trillions were lost due to exactly the same opacity. Everyone assumed building more and bigger datacenters, and digging to lay more fibre networks was necessary and where the value was at. They were right about the former, not the latter.

The picks and shovels investment thesis continues to excite to a frenzy, but we must only look to the darlings of the dot-com boom and take a deep breath - where are titans Cisco and Sun Microsystems today?

Demand continues to grow, however what we are willing to pay for it does not, and eventually the crop of businesses sucking in capital will need to show that it is sustainable and delivering value to customers. The jury is still out.