08 Oct
Well, we got here. Everyone’s wondering when a robot will take their job. Every pitch deck’s slide 3 is showing how they’re “AI Powered” with a vague flow chart of a neural net — whether it’s an AI business or not. And Elon Musk is already beginning the Battle With The Machines.
But all of this discussion is happening with very little regard to the actual state of AI technology. Skynet is not about to become self-aware anytime soon.
Yes, the advances in AI and machine learning are groundbreaking – but they’re in very specific domains. Areas such as NLP, image recognition, and voice have improved dramatically in recent years.
But the case for a reality check in AI start-ups is 3-fold:
1. The current breakthroughs still aren’t there yet
The formless superintelligence we fear still confuses chihuahuas and muffins about 10% of the time. And even when image recognition works great – humans can still completely break it by adding a little bit of noise in the picture.
2. Even the current breakthrough technologies are operating in very specific domains (ie, supervised learning)
Current AI technologies are made possible by massive amounts of training data and very smart supervised learning – basically, a well-defined goal to seek for. For example, the ImageNet database provides a clear and pre-defined answer to the question “Is this a picture of a cat?” Unsupervised learning, designed to work with much less structured datasets, has been on the horizon for years now but has had little success to-date.
For us to move to more generalized applications, we either require a much more sophisticated unsupervised model, or many datasets to train supervised models. These will likely both come, but both will take time.
3. Even if progress exceeds expectations, there’s still no way we can live up to the hype.
We’re very positive on machine learning’s long-term potential. In the long-term, we’ll see it transform the world in many profound ways. But at this level of attention, near-term expectations are being dramatically inflated.
Have no doubt – we ARE in store for a trough of disillusionment. A hype peak this high will inevitably disappoint some people.
So what should I do – as an AI entrepreneur?
So in short, AI has exactly the kind of transformative power that fuels great tech startups, but you’ll need to make it through years of non-plussed markets.
- Raise fast – You won’t have the power of the hype cycle on your side forever. Make hay while the sun shines.
- Have a real business – The best thing you can do in any market is build a lasting, commercially viable business. Don’t get pulled into the hype, keep your head down, and keep building.
- Utilize proven technologies – let the smart folks in academia work on the unproven stuff while you build a solid foundation with convolutional neural networks and LSTM.
And what should I do – as an investor?
- Focus on the next-gen of AI technologies – Unlike the entrepreneurs, your job is to see what’s on the horizon. Right now, the state-of-the-art is in reinforcement learning and adversarial networks.
- Avoid AI solutions looking for a problem – Generally, be skeptical of products that are only compelling if an embedded AI delivers superhuman results. Every great investment starts with a commercial case first.
- Stay domain specific – Some of the biggest successes to-date in AI are those who have either mastered a single domain (Clarif.ai, x.ai) or are building tools to support specific data science techniques (RapidMiner, Enigma, Trifacta).
Lastly, don’t get us wrong – we strongly believe that AI and machine learning have more potential for disruption than any other technology today, and will enable decades of productive innovation. We just expect these developments will take longer than the media expects. But the sooner everyone can get past the hype, the faster we can get down to building lasting AI businesses.
For more information contact us at info@oxfordvp.com