Inside | Real news, curated by real humans
Inside AI Missing

Inside AI (Jan 5th, 2020)

Happy Sunday and welcome to the weekend edition of InsideAI.  If you like it, please send this to a friend so they can subscribe.  I'm Rob May, a Partner at PJC, where I focus on Machine Intelligence.  I'm also a former hardware engineer and 2x venture backed CEO.  This week, I'm kicking off 2020 with some thoughts on where AI is going and in particular where I think the best opportunities lie.

There are a few high level trends that are really interested in the machine intelligence space.  The first is that AI is permeating everything and becoming a very horizontal technology.  Every company will need it, but at the moment, not every company has it.  It will touch almost every part of every business, but often in ways that we don't fully grasp just yet.  This is why it is so important to stay on the front edge of AI.  Much like the Internet brought new types of opportunities and new ways of thinking about work, so is/will AI.

The second trend that is interesting is that the limits of neural networks are becoming obvious.  While neural nets have a very long road of progress in front of them, the AI ecosystem is also starting to realize they can't solve some types of problems, and we may need new approaches.  This is exciting and will drive investment into other ideas.

The third thing that is changing is the vast increase in AI compute workloads is changing hardware and computer architecture in innovative ways and we have no idea what things will look like on the other side.  Hardware requirements vary across dimensions like size of footprint, ease/complexity of fabrication, power consumption, speed, heat dissipation, and ease of programming.  These are things system designers haven't thought as much about in years because the progress on all fronts, and the breakdown of options in the ecosystem, have been linear and standard.  All of that is changing because of AI workloads and the hardware innovation required to support them.

And finally, data needs are changing.  Because most AI models need so much data, we are seeing an early rise in two areas to deal with this.  The first is synthetic data - using a small real data set to generate a more robust and fuller data set to train a model.  The second is small data AI - techniques like one shot learning and probabilistic programming that allow models to be trained with less data.

So what does all of this mean for startups, entrepreneurs, and investors in AI?  I'll write more detailed posts about each of these over the year but here are my top theses for 2020.  If you are building a company in one, please drop me a note.

1.  Services as Software - This is rapidly becoming one of my favorite business models.  Take a "work stack" and keep the same human interface, but strip out everything behind it and use AI automation.  My investment that most symbolizes this is Botkeeper.  They pioneered some of the ideas of this model, and it has some interesting characteristics.  One of them is that, for some use cases, where customer acquisition is expensive because of switching costs, company acquisition makes more sense.  Automation in the services industry gives you the operational scalability and financial margins to grow via company acquisition in industries you couldn't before, and is often cheaper than you can acquire customers via normal marketing channels.

2.  Robotics that look like Software - As robot parts have become more standardized, what used to be a difficult industry to invest in is now looking more, economically, like software companies.  I've invested in 2 robot companies (undisclosed) that use a lot of off the shelf parts and whose innovation is in software and model training.  In some cases the economics look almost like SaaS.

3.  Non-Neural Network AI - I have yet to make an investment here but really really want to as I think their are big opportunities.  It could be a full stack approach, biting off a major problem with a non-neural net tech stack, a cognitive architecture approach that uses some neural nets, or a platform approach that provides non-neural net functionality to others. (Though the latter is a tougher sell at the moment)

4.  AI Hardware - As I have written many times before, I'm a big fan of the ecosystem explosion in AI hardware, and have already invested in Mythic, Rain, and Koniku.  As the market starts to mature and use cases become clearer and somewhat standardized, I expect there are lots of opportunites in this space.

These are the more general trends where I see opportunities.  I also see a lot of very specific application opportunities, some of which I will write about in coming weeks.  2020 should be a great year for machine intelligence overall, so I hope you will keep reading InsideAI, and keep sending me your ideas and questions.  Happy New Year.

@robmay