The First Wave of Corporate AI Is Doomed to Fail | Hacker News Reader

The First Wave of Corporate AI Is Doomed to Fail

hbr.org

TLDR "AI will transform industries, it hasn't happened yet, some companies have given up because it's too hard, some companies haven't, mobile apps are cheap and so too will AI be."

This is the best business model for AI startups right now: Raise money (at least $10m) to start a private equity fund. Now find businesses that YOU know will massively benefit from AI but their management simply doesn't understand. Buy the company. Use AI to increase margins. Sell the company. Profit.

This is a massively better idea than selling your time for X to companies run by people who don't understand what you'd be doing for them.

Actually, there's a bit more to it, but that's where the next wave of AI profits will come from (PE funds, not startups and not VC).

Speaking of which, if anyone is doing something similar, feel free to contact me.

That's a great idea.

The subset of people who can raise >$10m for a private equity fund and that understand current AI is probably quite small.

It's a pretty open strategy at this point. In a few years, everyone will be doing it.

Also, PE itself is such a messed up industry. Very conservative, hasn't changed their playbook in 2 decades. The guys at KKR aren't the ones that are going to figure this out.

Maybe a PE expert should partner with an AI expert. They seem rather different skill sets.

The only problem is this is almost every non-tech company. I guess you could focus on companies where the core business can easily be improved by an algorithm. But there's opportunity everywhere.

Except: early AI has already succeeded. Things like Siri and Alexa work everyday for millions of people. Then there's also the hidden AIs we so ingrained in our life we don't notice like Google's searching or even credit card fraud detection.

The AI industry has a terrible problem: once we create an algorithm for something it's no longer AI, it's just an algorithm.

It feels like there should be a corollary to Arthur C Clarke's famous law: Any sufficiently advanced algorithm is indistinguishable from intelligence.

Excellent! Couldn't agree with you more

Or "Any sufficiently understood tech buzzword becomes distinguishable as an incremental improvement on preexisting tech"

At this point, it's clear that the only version of AI that'll be acknowledged by the hard skeptics is "a system that thinks and works exactly like the best humans in history along every single individual axis, except a kabillion times better."

Well, that's what AI is always hyped as, by its proponents, and its detractors who want us to beware of a singularity where godlike machine intelligences take all the jobs and then turn the entire universe into paper clips.

After being fed a diet of daily articles about the immense power of AI, it's pretty natural for someone to look at a program which gets your spoken song and movie requests right 75% of the time and say "that's not AI".

> Once we create an algorithm for something it's no longer AI, it's just an algorithm.

Yes. If we create an algorithm, what intelligence is involved: real or artificial?

Even using neural networks we're a long way from not having to give a high-level scaffold for an algorithm we've already decided upon. Present day neural networks are essentially a glorified spreadsheet "goal seek" function.

An apparently smart image recognition "AI" has really only distilled the intelligence of the humans that classified the training data, and generalized it to other images in a manner devoid of real intelligence, which is why image recognizers can report that specially-prepared TV static is 99% certain to be a peacock.

You're conflating voice-interface with a.i. they're not the same thing, they just appear to be the same thing to the untrained eye.

A machine that could understand natural language used to be considered AI. It's more the definition of AI that shifts.

I actually think this misses the biggest reason that this AI cycle could fail. A great deal of machine learning talent is concentrated in big companies right now, because of the nice salaries and stability it provides research initiatives.

But the real disruptive applications of AI will necessarily apply to new, untapped markets. It won't fit into search or ad markets, for instance.

Until AI moves out of research at labs and bigco into the startup space, it will be necessarily limited in scope.

Interesting because it could be the failure of this cycle that 'releases' talent into a startup ecosystem that lays the seeds for the distruptive next phase.

> Until AI moves out of research at labs and bigco into the startup space

Sounds like what OpenAI (https://openai.com/about/) is trying to achieve.

Somewhat. As I see it, openAI is trying to build the tools (and research papers, IP, etc.) in a publicly facing venue that could power the frameworks that ultimately power the disruptive wave of startups.

IMO, there's almost too much focus on the really sexy research areas of "AI". Everyone wants to work on self-driving cars and chat bots. But there's huge opportunity in just applying basic models to business processes with normal data.

This article reads as rather lazily-written. Consider:

Allstate car insurance already allows users to take photos of auto damage and settle their claims on a mobile app. Technology that’s been trained on photos from past claims can accurately estimate the extent of the damage and automate the whole process.

How do we make the jump from very well-established (even boring) use of tech--sharing photos--to computer vision and analysis that can value a vehicle, understand the damage both visible and concealed, and assign useful figures for both repair and value? And won't such results be disputed at an astonishing rate? I mean, sure, it's a nice idea, but there does not seem to be any bridge at all from one concept to the other.

The good news is that emerging marketplaces for AI algorithms and datasets, such as Algorithmia and the Google-owned Kaggle, coupled with scalable, cloud-based infrastructure that is custom-built for artificial intelligence, are lowering barriers.

Renting someone else's algorithm sounds like a very risky proposition, as they can easily drive you out of the market when it becomes worthwhile to do so (see also Amazon's strategy around fungible third-party items that are profitable).

Was it just me, or did the title have little to do with the actual article?

Everything written except the title seems pretty reasonable to me, and I'm an AI optimist.

Yeah, those spam filters sure are a big failure all right.

Reply on news.ycombinator.com