Blog: Electricity, Cheap AI, and the role of the new Data Scientist.
Recently I had a conversation with a manager in a non-tech business about how Artificial Intelligence is on the verge of having big impacts on all aspects of the economy.
He was looking for a data scientist to help make predictions to drive upcoming business decisions. This got me wondering how to understand the role of upcoming AI in business better.
In this article, I’ll try to answer this.
Artificial Intelligence as the new electricity
As cobwebby as it sounds this analogy has are some very useful parallels.
For as long as people have existed, they have been awe-struck by lightning. Some attributed it to the wrath of a God, while some others wondered how to put it use.
Between the gritty lightning probes by Benjamin Franklin’s kite in the 1700s to Edisons revolutionary and commercial light-bulb in the 19th Century, this ‘magical’ and ‘dangerous’ substance hardly had any substantial impact on human life.
Even until much later on, electricity was referred to as ‘light’ and the new and dangerous inventions like the ‘murderous’ washing machine that utilized electricity were all plugged into light sockets. (watch Jeff Bezos talk about washing machines).
It wasn’t until much later that electric lines were laid out and the standard socket was invented. Then, in 1935, Franklin D. Roosevelt led the creation of the Rural Electric Administration to bring electricity to rural areas identifying the essence of its utility, and the rest is history.
So what changed between the conception of electricity and the wide-scaled usage?
- Real World Utility (brought by the plethora of electrical appliances)
- Drop in price.
In 1850 when a politician Michael Faraday was asked why he thought electricity was valuable, he answered, “One day sir, you may tax it”.
Just as the price drop and the increase in the quantity and quality of applications helped electricity transition from a searing new invention to an everyday necessity, so will the phenomenon of cheaper artificial intelligence and the coming onslaught of innovative AI applications change businesses and everyday lives.
What does Artificial Intelligence do, really?
Some think that researchers gave up on the pipe-dream of Artificial Intelligence with the last AI ‘winter’. Others think that now is the AI ‘spring’. Perhaps we are on the verge of a scorching AI summer (with cryptocurrency induced heat waves 🤣)
I think that, the seasons analogy is a tad bit unnecessary. What has actually happened is that the previous notion of a human-like AI has been usurped by the data revolution, and it seems rightly so.
In summary, the data revolution meant that: 1) For AI to be useful, it does not require common-sense intelligence at all 2) Training an AI system can be done end-to-end using only data (hierarchical representation learning or ‘deep’ learning)
While science may still be interested in a Turing-Test-passing, empathetic AI that demands citizenship, research towards this has been mostly pushed under a new umbrella (see Artificial General Intelligence) and has far less enthusiasts as does the data driven approach.
Right now, the majority of the work seems to geared towards: 1) AI that can perform as well or better than humans at specific tasks 2) ensuring that this ‘performance’ has utility in our lives. (everyday applications)
How is AI useful, for you? Cheap predictions
At the core of AI are predictions . All flavors of machine learning including reinforcement learning deal with how to predict the unknown by inferring patterns from the known. The unseen in some cases is a ‘type’ as in the case of classification, a ‘value’ as in the case of regression or a policy to drive further actions (reinforcement learning).
However, predictions in and of themselves aren’t very useful if they aren’t followed by decisions. For instance, it is far more useful if I can take the prediction from a machine about a potential employee that is about to quit and be able to intervene and prevent that (decision and action).
Then there is outcome. Stretching the HR analogy of a potential turnover, it might not always be a good idea to try to incentivize someone who wants to leave to stay. For instance, they may stay but not perform, in which case the outcome doesn’t justify the decision.
After the decision, there is feedback. The continued performance of employees who were convinced to stay after they ‘almost’ quit, can be tracked and analyzed and used as feedback to make the prediction and decisions processes better.
Machines are better at the task of most predictions than humans. (and if they are not, they will be). This is because the newer architectures are very specialized (only one task) and once trained, they are more consistent (same input produced same output, no fatigue or issues with loyalty) and are much more affordable (automation) than humans.
This will lead to an increase in adoption of AI techniques to make predictions . This increase in turn, will lead to even more decisions, in two key ways:
- more decision-makers can make informed decisions (previously restricted due to lack of informed predictions), and
- more decisions will actually be made by existing decision makers (as opposed to left to the default due to lack of information.)
Let me clarify with a little story. I have an uncle in LA who takes great pride in having lived there so long that he knows all the streets and can surmise traffic conditions to take the most efficient route. This ability must have been of great use in the past, but now with traffic aware routing available and free!, I can navigate LA just as well, probably better (satellites!).
The consequence? 15 years ago, I would have probably asked him to come pick me up at the airport, now, I can confidently rent a car and make the driving decisions on my own.
Similarly in business, AI will lead to cheap (almost free) predictions and thus it will level the playing field so that there will be many more decision makers in the playing field. It also means that existing decision makers, will make more decisions and leave less to default.
What does this mean to businesses that rely on having propriety industry knowledge they gained from years of practice? OR, those whose expertise is in providing these predictions at a hefty cost? Time will tell.
The dwindling demand for the expertise of yellow cab drivers may be an indication.
Accurate ✔️, cheap✔️ but understandable?
Humans have traditionally made predictions and decisions simultaneously.
Indeed, separating the two is considered the very penchant of a rational being, and works in cognitive psychology have highlighted the dangers of not doing so. It is not surprising that monarchs in the past would have advisers whose sole job was to offer predictions and bias checks.
Notwithstanding, humans if left to their elements tend not to make the most rational choices (we are naturally poor in statistics). This is the reason that statisticians (and other experts) were trained to be aware of their own biases and help decision makers overcome this shortcoming.
However, with complex and non-linear (but highly accurate) prediction techniques like deep neural networks, the traditional expert is becoming less handy. In fact, even the developers of these systems (like me!) consider them to be ‘black-box’ — we know that the system does very well, but we don’t know exactly what it is doing, or how.
Role of the new data scientist
This is where the newly minted field of data science can shine, perhaps.
The job description for a data scientist, depending on the size and the pocket-depth of a company, is advertised to include some or all of:
- Collect Data (Instrumentation, Logging, sensors, user-content)
- Move/Store (ETL, Data Flow, Pipelines, Horizontal and vertical expansion)
- Explore (Clean, anonymize, detect anomalies, cluster, sample, prepare)
- Aggregate (Analytics, Metrics, Segments, Engineer useful features)
- Learn/ Optimize (A/B testing, Experiments, Simple Machine Learning, Deep Learning)
What company executives are looking for is someone who can ‘wrangle’ the data and present it, usually in the form of visualizations such as the one below. This is so that they can make more informed decisions. The hope is that as insights such as this are cheaper to produce, the quantity and quality of business decisions will improve.
Visualizations such as this and many others can now be done quickly and more efficiently thanks to the data visualization ecosystem. Moreover, with drag and drop software like Tableau, they can also be done with minimal coding (to a certain effectiveness).
However, there are some caveats. Almost all real-world data has many factors (the figure above has one- number of tweets across time) that make visualizations less intuitive.
Further, the relationships are almost never cleanly linear (i.e. there are many factors that combine to produce a result). This is why Deep Learning that combines these factors non-linearly and hierarchically usually has performance gains but leaves us with questions like:
- How should explain the prediction process of these complex models to decision makers when the developers themselves are not quite sure?
- How to take a deep neural network with 21 million knobs, and come up with a reasonable explanation that it learned the ‘right things’ ?
- How to analyze that if the data would have been different in a certain way, the results could have been different in a particular way (the counterfactual)?
For that, I add a new task here for the data scientist of today:
6. Explanation: explain predictions truthfully and in a user understandable way
The above figure is a visualization of one of the layers (hierarchical representation of input features) to give an idea on what the neural network is ‘seeing’ when it looks at an image of a dog and a cat.
Visualizations such as this aren’t merely about wrangling some data and using the latest plotting library. They provide a window into the workings of a neural network and its learning mechanism and take us further into answering some of the questions above.
In this previous article, I talked a bit about explanations. In later articles I will write about what it means for explanations to be truthful and human understandable.
Rebooting Strategy: More Automation
I mentioned earlier that the feedback from outcomes can be tracked and fed back to make the prediction process better, this is the essence of what is called active learning.
The same feedback (if plentiful) can be also used to improve the decision making itself as well. The permission to make mistakes and then to learn from these mistakes is the essence of reinforcement learning. This is probably what we mean when we say a certain decision maker has experience.
If the successes in applying reinforcement learning to complex processes (alpha zero, complex games and applications) is any indication, there is no reason why AI in the near future cannot be used to learn better strategies for making decisions as well. Doing this will not only affect the quality and types of decisions being made, but will also fundamentally change the strategy and structure of businesses — all that for a later post.