The title of the post is perhaps a little misleading in one way, I really don’t think Elon Musk coded his AI, but I do think both he and George Hotz set the features, relationship and prioritisation around the work for their AI’s and in Georges case also built it.
In thinking about the design an artificial intelligence too many people think about the inputs and not the logic first. The critical path in AI design is
what are the features of the AI that make it intelligent beyond a binary interaction and where are these interactions focused
Karl Smith, Founder at UbiNET
Geohot took the classic hacker approach in relation to immediacy in defining his features, taking the problem statement to ‘how to teach a car to drive’ instead of ‘how to understand the world in which a car moves’. These are fundamentally different approaches to the same question producing radically different solutions.
Tesla and Elon Musk
The Tesla solution is an ecosystem approach, from friend who have them the end to end experience of buying, getting and using a Tesla is orchestrated in much the same logic as the AI is designed. The Tesla AI maps the world and creates not just an in the present experience but a sense of knowing relative environment and future risk. This is much the same way as drivers make constant assessment of context based factors in their driving responses. However this way of building AI is extraordinary expensive not just in the initial build but also in the maintenance with constant updates.
The George Hotz solution is at the other end of the spectrum, its focusing on actually learning to drive and making the AI responsive to environmental change rather that mapping the world. In the film below George makes a statement
Drive naturally like a human not some engineers idea of safety
at 4.50 onwards which I used in 2016 when I spoke at SXSW in Austin, USA about Cognition Clash in the Internet of Things, that is fundamental in building Artificial Intelligence that is adaptive rather than limited by human perceptions on how we think we do things.
Clients don’t understand their customers, they just think they do!
It’s not for the lack of trying or spending millions on developing and building huge data systems, the problems are many but can be traced back to one simple thing;
“Data only describes part of the what is happening and almost nothing of the why, let alone what should be done to change the situation”
Clients have been sold that data gives them the answers and that big data will close the loop for them to understand the upstream and downstream thinking of their customers, WRONG.
Douglas Adams noticed the real problem
Douglas Adam’s said “But even Amazon has only got part of the picture. Like real world shops, they can only record the sales they actually make. What about the sales they don’t make and don’t know that they haven’t made because they haven’t made them?” Douglas Adams “The Salmon of Doubt” by Permission of Pan Macmillan. That pretty much covers the problem if you extrapolate the thinking for Data Analytics, Big Data or even Artificial Intelligence based Data and Decision systems.
“Data is binary a yes or no (even complex views), it does not capture motivation, intention, desire, cognition, distraction or any other human reasoning or pattern”
A child pretending to be a robot just as data pretends to be the truth, he is a kind of robot and data is a kind of truth
A pure Data approach to understanding customers will provide the wrong data because data is an absolute and people are not. Even with Artificial Intelligence it only works from the starting point you give it, if any of the perimeters are wrong the whole data sample is wrong.
Guide to understanding Customers
Data, Big Data, Artificial Intelligence – Tells you what
People in target demographic – Tell you why
People in target demographic
User research answers the question Why have we not made the Sale? through the only people equipped to answer the question, consumers. This is not market research, its scientific without a predetermined agenda or outcome. User Research is a problem solving method that offers solutions by finding the right questions, finding the right people and asking the questions in a way that does not lead or direct the answers.
There are right questions and people to ask?
This may sound a little Adamsesque (if you ask the answer to Life, the Universe and Everything you get 42, because it the wrong question). Getting the questions or setup wrong is the real problem with an Analytics approach to a Diagnostic process. While it may be reasonably expected by a seller to directly ask, why didn’t a visitor become a buyer or register. Visitors may be asking themselves where am I? what does this do? this does not make sense, should that be happening? technology, why do I bother? Why has my screen gone pink? None of these “in mind” experiences are expressed in the data or even a consideration for the data schema design.
A visitors experience is not only defined by the online environment but they bring past experiences, desires and doubts about their current experience. Without these insights from research, it is difficult for clients to grasp potential problems, gain a good return upon their investment (ROI), innovate to fit the market and consumer needs or break into a new market sector.
Reasons that Data is Trusted and People are Not
It appears to come down to scale and a short sighted approach to costs. Buying an Analytics Solution appears to tick all the boxes, even if in reality it does not. While using Research Companies or in-house Research Teams seems expensive in comparison.
“The real trick is to understand you need both, you always did”
When I first started using Web Position Gold (the analytics tool), bought by Webtrends long before Google Analytics existed or the current proliferation of products promising the impossible, we used it to spot trouble only. We would then do some user testing in the area, working out possible failure scenarios, from there we would suggest two or three solutions and build them for A/B testing to see what worked and what did not. Everything was monitored and all the data from both analytics and user testing was collated into one final solution. Sometimes there was a single resolution, a re-architecting of a section, in one project I kept 16 pathways active because they all delivered transactions for different types of customers.
The thing is just as there is no absolute way to find out the problem, resolution or adaptive innovation except byDiagnostics a digital and human activity.