The AI conundrum — is it worth it?
The most interesting thing I read this week was Routes to Defensibility For Your A.I. Startup.
It’s a great analysis of places where incumbents have an advantage, and where startups might be able to compete when it comes to A.I. It also looks at strategies for entering markets with A.I. approaches, and how to develop the moats of the future. As you can imagine, it is heavily focused on data types, and data networks effects, which do indeed drive most of the advantages in A.I. at the moment. I do have some thoughts that this type of product focused defensibility might not be the only option though.
A framework for understanding data network effects and incumbents’ advantagesmachinelearnings.co
The big idea article above talks about product related competitive advantages in A.I., and this is where most analysis has focused. Products that have access to proprietary data perform better, and thus get more customers and more data. It’s true, but I want to add a layer to that and offer a more complex relationship as an option as well, based on what I’ve seen in the A.I. market so far. I think that understanding how to sell A.I. could be an early, temporary advantage that could setup a company very well long term.
From what I can tell, most companies selling A.I. are actually struggling to do so, for several reasons. First of all, it can be a lot of work to implement a new A.I. system that only does one small new thing. Because most A.I. products are trained on one data set and one model, if you think of how they apply to the business world, they are often niche mono-products, rather than fully functioned products that can do a lot. So you have to find these targeted use cases where doing A.I. on one type of data is useful, because, in many use cases, you would need to to apply A.I. across several data sets and work functions to make it useful enough to sell.
The second issue is that deployment, particularly getting access to the customer’s data to train models (if you need it) can be a point that slows down the sales cycle. I see most A.I. companies doing paid pilots, which isn’t something I saw as much of in the early days of SaaS. But, buyers are worried about how to implement A.I., and how well it will really work.
The third issue is that buyers don’t understand the probabilistic nature of many A.I. systems and how to integrate them into their workflows. With SaaS, when you click button X, action X is performed 100% of the time. With A.I., you may get an outcome that is an estimate — say 96%. What does that mean? Is it wrong 4% of the time? And if so, how do you know when to ignore the machine? And more importantly, who in your company is responsible for this given that A.I. isn’t usually someone’s job at the moment. I’ve seen many of my angel investments struggle with this last point — that sometimes buyers don’t know who should own the A.I. software workflow in their companies.
From my perspective, figuring out how to sell A.I. at scale is still a challenge for most companies, given the current state of the ecosystem. Being an early company to solve that problem may provide you with quite an advantage in data acquisition and market perception as the leader, even if you don’t start with the most data or have the best algorithms. It’s a really good place to try to make an initial stand.
How To Spot A Machine Learning Opportunity, Even if You Aren’t a Data Scientist.
Kathryn Hume points out that having a general understanding of how these A.I. algorithms work, is becoming an important business skill, and provides some ideas for how to find opportunities in your own work.
Artificial intelligence is no longer just a niche subfield of computer science. Tech giants have been using AI for…hbr.org
The Shape of Work To Come
An interesting look at how A.I. and other digital technologies are shaping the future of work.
Illustration by Chris Malbon Last year, entrepreneur Sebastian Thrun set out to augment his sales force with artificial…www.nature.com
How Machine Learning Is Helping Neuroscientists Understand the Brain.
Neuroscience oftens inspires developments in A.I., but can the reverse inspiration also hold true?
The workings of the brain are the greatest mystery in science. Unlike our models of physics, strong enough to predict…massivesci.com
The Machine Intelligence Continuum
What types of machine intelligence systems are out there, and what can you use them for? This piece is a great look at the various key technologies and the types of systems they can be used to build.
This is part two of our WTF IS AI?! series. Read part one on modern AI techniques if you missed it. If you're not an AI…www.topbots.com
Other interesting links:
What Happens When Machines Know More Than People Do?
One of the most controversial psychological studies in recent memory appeared last month as an advance release of a…www.strategy-business.com
Growing the Artificial Intelligence Industry In The UK.
This independent review, carried out by Professor Dame Wendy Hall and Jérôme Pesenti reports on how the Artificial…www.gov.uk
DeepMind’s Go-playing AI doesn’t need human help to beat us anymore.
Google's AI subsidiary DeepMind has unveiled the latest version of its Go-playing software, AlphaGo Zero. The new…www.theverge.com
Google’s machine learning software has learned to replicate itself
Back in May, Google revealed its AutoML project; artificial intelligence (AI) designed to help them create other AIs…www.sciencealert.com
How do we bring back manufacturing jobs? Hire more robots
For decades we’ve been told robots were to blame for the dearth of manufacturing jobs in the US, but that’s about to…medium.com