AI as Kinetic Energy: The Creation of New Categories
- Feel free to leave comments. I’m fairly reachable so if you’d like to connect please reach out as I’d love to hear your ideas- email@example.com :).
- I’m going to break this into 3 or 4 waves since I’d like you all to actually read it. As such, we are going to get through this together, but it’s going to take a bit longer!
- Buckle up and keep your hands and arms inside the vehicle at all times.
Imagine for a second a snapshot of a carpenter in mid-swing with a hammer at the very tippy top of it’s arc. At this point in the carpenter’s action the potential energy is at its highest, and all the hopes and dreams of that hammer rest in this climatic moment. The system is in perfect suspension waiting for something, anything, to catalyze motion. Once that system is triggered it’s very hard to stop it from converting into kinetic energy before reaching it’s second climax right before it smashes the nail below. AI is not new technology, and after 50+ years of development as a discipline, I’m not sure what served as the final catalyst for this rise in AI (better hardware, cheaper compute power, etc.), but I know for certain at some point in the last 18 months or so that the hammer started moving, and in my opinion we shouldn’t be fighting it, but rather using that momentum to our advantage.
For me the epiphany came when 9/10 pitches I was seeing had some AI component to it. It’s like they always say if everything is AI, nothing is AI. I’ve seen enough .ai companies claiming AI, that I’ve started to develop a loose personal litmus test to suss out the wheat from the shaft. Quick story, I was grabbing a beer a little while back with a founder of an unnamed AI company, and she dropped a great line on me in saying that artificial intelligence is just that, artificial. Her premise behind the comment was that it’s the people behind the application of the technology that really make the difference between intelligence and not, which includes both the people who created it, and the people who are leveraging it.
From the perspective of early stage venture, what we debate often is how to build a big AI company. At Wildcat, we believe we have some good examples in our portfolio companies Workfusion, Amplero, and others, but the vast majority of the startups we see are going to cap out as features never making the jump to products, platforms, or companies. Many of these feature companies have the capability of getting acqui-hired in the $50-$250m range, which is certainly nothing to scoff at, but they aren’t a big enough opportunity for us to be excited about them as investments.
One of the biggest limiting factors in getting past the feature stage is taking the technology beyond that of a project. What I’ve observed for pure AI companies, and this could be isolated to my experience, is to pick a problem big enough that can scale across multiple industries and establishing a clear go to market strategy to get you there. As one of my colleagues called it, figuring out how to move from a “vertical” technology to a “segment vertical.”
There is a fine balance making that transition because a lot of these AI/ML applications are limited by their data. The implication for these startups is that they have to acquire a proprietary dataset that is broad enough to reach across multiple industries, but has to solve each problem it tackles very well. This all has to be done while balancing what the big boys have. Let’s take the example of facial recognition, as paraphrased by Andrew Ng during his WSJ CIO Keynote, a startup is going to have a really hard time going up against Baidu’s 200m face dataset when the largest publicly available one is 15m. I’m looking right at you 98% of early stage CV companies, pick more creative applications! Check out Kairos’ list for best face databases and you’ll see how behind your startup really is.
When entrepreneurs have big ideas going after multiple big industries, sometimes they come off as too scattered when presenting the capabilities of their technology if their go to market strategy isn’t crystal clear. I liken this to walking into a room full of puppies and having a hard time picking which one to play with first. These companies never have a shot of getting ‘there’ no matter how impressive the technology because they’ll get too distracted by the shiny objects along the way- hint- you come off like the golden retriever in this video. A more successful tact I’ve observed is someone attempting to build out a big business that uses AI as a core element of their overall solution.
Another no less important part of this balance is the creation of a category. At some point people lose interest in incremental technologies, and the companies that get really big build something no one else touches and dominate an entire market of previously trapped value unlocked by their product.
Wildcat is making a concerted effort to seek out category defining companies that, as my colleague Geoffrey Moore would say, take something that was scarce and expensive and make it ubiquitous and free. In this vein one of our hypotheses for AI is that it will create very different looking economy where people can move away from unstimulating repetitive jobs and manual labor to an economy enabling creative problem solving to be a bigger part of more people’s day to day life.
This series on AI will focus on what that looks like in more detail; I hope you stick around until the end.
Thanks for reading,