Burning Questions on the Future of AI

When we ask the question “What’s Next in Computing?”, artificial intelligence is arguably the most exciting area of development. There has been a rapid progression in the space, and many believe we are now entering into the golden age of AI. This progress has been made possible through three significant recent breakthroughs: cheap parallel computation, big data and better algorithms.

With its recent explosion, I have become very excited about AI. I began to consume information about the space daily, which was only to realize that we are in the very early stages of AI and that there are still endless questions to be answered. In this post, I will pose my most pressing questions about the space, and lay out my thinking about potential answers to these questions.

Some of the questions I have:

1. Where do the most promising opportunities lie for startups?
2. What will be our third run-time as a consequence of recent developments in AI? Is it messaging, maps, voice assistants, bots or push notifications? And what are the potential use cases of each?
3. And then there are virtual assistants. How are we going to interact with them? They are a highly relevant issue as the line between our personal and professional lives gets blurred with the rapid introduction of virtual assistants.
4. Let’s focus on enterprise for a bit. What is ripe for automation within our regular work processes? I used the word “work processes” for a reason as I believe AI (or machines) will replace or change work processes rather than entire jobs — at least for a very long time.

I will use these four questions as the foundational framework for this mini series, and I will cover one question per post. Today, I will start off by tackling my first question on potential opportunities for startups in the AI space. Here I’ll discuss about my approach to analyzing the AI space based upon what I’ve read and interviews I’ve given to some industry insiders. Needless to say — this is not an exhaustive list.

Companies Liberate Basic Development of AI Applications

AI has long been the domain of academia — academics, were ones who stayed with the field during all those long AI winters, after all. There is a gap between academic research and potential AI applications, and we need startups that work on bridging that gap by offering dead simple API- based solutions. We need solutions that will liberate basic development of AI applications. This is what Ilya Sukhar likes to call: “Parse for AI”. He argues that we need companies that come up with “3–4 well scoped features” and “someone who primarily writes frontend Swift/Java should find value in API in <10 mins”. I couldn’t agree more.

Clarifai is a great example. It offers image classification and visual search capabilities to other companies and developers. It makes AI “available for everyone” through its best in class API solution.

Other successful examples are Textio and SigOpt. Textio has a text analytics and data visualization software. SigOpt provides optimization algorithms for enterprise. It’s worth noting that these are both SaaS solutions.

There are many other applications such as voice recognition and text categorization that startups can differentiate by offering simple, API based solutions to democratize the AI space.

Special Purpose Tools Enable Non-Technical End Users

In the previous section I described how to enable developers, and now I’ll talk about everyone else, namely, non-technical end users.

One area of application is offering tools that allow non-programmer end users to perform sophisticated data analysis. There are promising startups active in this space such as Dataiku, Paxata and Context Relevant. Dataiku’s Data Science Studio allows people working on data to clean data, build models on top of that data and bake those models into existing work processes. However, you still need to be “data proficient” in order to be able to play with data. Paxata enables analysts to clean up data quickly and make it ready for analysis. However, I strongly believe that Slack of sophisticated data analysis has not been created yet, and we need this. We still do not have a simple and clean set of data analysis tools for everyone’s use in the enterprise. Current AI/machine intelligence applications coming from startups focus too much on developers and data scientists, applications democratizing AI for end users will be the next interesting wave.

Companies with Hard to Replicate Datasets

Josh Nussbaum argues:

“It is likely that over the long term algorithms will become a commodity. Since the real value then is in the proprietary data set collected, a startup is at a disadvantage on day 1. This is where first mover advantage actually matters. As a startup collects the data necessary to feed their ML algorithms, the value the product/service provides improves, allowing them to access more customers/users that provide more data and so on and so forth.”

I agree. It’s not the algorithms that will differentiate startups in the long run, it’s the access to proprietary data, and I think there is a play for startups that make high quality and focused data accessible to others.

A great example is Foursquare. They use their two popular mobile apps to collect invaluable data from consumers, package the raw data into multiple data products, and offer these products to businesses in the form of a freemium API.

Some other successful examples in this space are Factual, Clearbit, Planet Labs, Enlitic, Premise, Truven Health. However, this is not an easy play. Ideally, startups should also offer solutions to business problems from the companies from which they collect data. But once proprietary data has been collected, startups will have a strong competitive advantage, and there are still many untapped verticals to explore.

Agents

Many people spend a significant amount of their days doing some form of administrative tasks. There are multiple tasks in our everyday lives that are ripe for automation with the advancements in AI.

Enter bots. Because there is a certain hype attached to current discussions about bots, it is hard to differentiate signal from the noise. However, there is a great opportunity for bots that solve very specific user needs. And we are still in the very early days in terms of mass adoption of “agents”.

Some examples include x.ai, Operator, Facebook M, Howdy.

Conclusion

AI/machine intelligence is a very active and exciting space. Big companies are competing in the race to acquire the most promising AI startups. I listed four investment areas that I will continue to follow closely. However, I don’t think these four areas are all equal in terms of their attractiveness for investment. I’m more excited about first two areas, because they solve fundamental business needs for enterprises and their business models are fairly proven.

Before concluding, let me recall the four-question framework I will use for this mini series:

1. Where do the most promising opportunities lie for startups?
2. What will be our third run-time as a consequence of recent developments in AI?
3. And then there are virtual assistants. How are we going to interact with them?
4. Let’s focus on enterprise for a bit. What is ripe for automation within our regular work processes?

In this post, I have tried to cover the first question. In next three posts to come, I will tackle remaining three questions separately.