Don’t be scared. AI gives humans Superpowers. A practical Retail vertical perspective.
“Any sufficiently advanced technology is indistinguishable from magic.”
This is known as the “Third Law” of the famous British science fiction writer Sir Arthur Clarke. To me, no other technology illustrates this better than Artificial Intelligence (AI). Let me explain why I believe that AI can be nothing short of magic.
It’s estimated that humans have been around for about 6 million years and the modern form of humans about 200,000 years. If you look back at humanity’s history until the 20th century, it has been quite boring from a technological and economical perspective. Yes, there have been some inventions like the hand axe, controlled fire and agriculture, but none of these really moved the needle and technological and economic progress was just incremental. As so well described by Brynjolfsson & McAfee in their book The Second Machine Age, until the 20th century there has only been one dramatic step change in progress.
This was in the late 18th century. Nearly simultaneous developments in mechanical engineering, chemistry and a couple of other disciplines, led to the invention of the steam engine developed by James Watt. This allowed us to overcome the limitations of human and animal muscle power and led to factories and mass production. In hindsight, we call this the Industrial Revolution.
Note that muscle power wasn’t just replaced. Rather, the advent of steam power in turn made a power multiple orders of magnitude larger available. More importantly, note that the Industrial Revolution didn’t lead to higher unemployment rates, but to people switching to other, more attractive and creative roles.
After this — from a macro perspective — technological innovation has been relatively incremental again for about 200 years.
But now there are again several technologies coming together which are kicking off the Second Machine Age. Advances in computing power and availability of data have reached a tipping point. This means that artificial intelligence models, that have been around for many decades, have suddenly become far more powerful. Combined with the internet, this is driving almost all the major innovations we see around us at this moment: driverless cars, voice recognition, computer vision, augmented reality and robotics. In this Second Machine Age, part of human brainpower is being replaced by artificial intelligence.
Again — like in the First Industrial Revolution — those parts of human brainpower aren’t just being replaced, but in narrow domains a level of intelligence multiple orders of magnitude higher is introduced. Again, I believe this Second Machine Age will not lead to large numbers of people becoming unemployed, but to people being relieved of the boring operational parts of their jobs and being able to focus more on the attractive and creative parts of their jobs, and more attractive new jobs, altogether.
But isn’t AI the technology that is characterised as an existential threat by the greatest minds of our time, such as Stephen Hawking and Elon Musk? Indeed, as explained by Oxford University philosopher Nick Bostrom in his book Superintelligence, there is significant danger in developing a “Strong AI” which is superintelligent — i.e. more intelligent than humans. Although in general I believe governments should intervene in markets as little as possible, this is clearly an area that needs to be highly regulated by governments and intergovernmental organisations, just like nuclear energy. This is of course very sensational to write about, which unfortunately has led to most of the press either covering AI as an existential threat or focusing on the fact that robots will take all our jobs.
However, all current forms of AI are “Weak AI”. This is defined as artificial intelligence that is focused on a narrow task. Although the speed of learning and skill in a narrow domain can be jaw-dropping — such as in this YouTube video of Deepmind learning to play the video game Breakout — this is different and far from strong AI. One might be misled by a recent publication by Google researchers titled One Model To Learn Them All. Although this neural net can be used in multiple domains, such as speech recognition, image classification and translation, this is still very far from artificial general intelligence. Nick Bostrom reports that most experts in the field estimate that we won’t reach human-level intelligence (HLI) before 2050 and that true superintelligence is likely to be reached 30 years after reaching HLI. Luckily this gives the human race some time to prepare for the arrival of Strong AI.
It is my strong conviction that powerful technologies cannot be stopped, but we can and should make sure they are used for good. In addition, I believe weak AI to not only be beneficial, but a necessary response to other developments in the world around us. I will share an example from retail, because that is the industry I know very well as Founder & CEO of software-as-a-service company Omnia Retail. However, I’m sure this will be true for many verticals.
Retail has been fundamentally disrupted by the internet. Where before the rise of the internet retailers’ assortments were limited by the physical shelf space available, in e-commerce assortments can in principle be unlimited. E-commerce and omnichannel retail (retail across “channels” like physical stores, desktop internet and mobile) have also led to increases in competition. Where retailers previously only had a couple of local competitors in the neighbourhood of their stores, now they often have at least 20 competitors offering the same product to the shopper’s home, including international tech giants like Amazon and Alibaba. Finally, there has been a fragmentation of marketing channels. Where in the past a retailer only had to worry about which products to promote in their weekly circular, nowadays a retailer’s full assortment — including prices — is visible 24/7 on channels like Google Shopping, comparison shopping engines like PriceGrabber and Shopzilla, banner ads showing products similar to the product that visitor has viewed before on the retailer’s website, and many more channels. These three drivers had led to an explosion in the number of pricing and online marketing decisions retailers need to make. Where in the pre-internet era an average retailer had to make about 4 thousand pricing and marketing decisions once a quarter in order to stay competitive, now an average retailer needs to make 60 million pricing and marketing decisions multiple times per day. This is a completely different level of complexity.
To cope with this increased complexity and explosion of data to process, automation and AI are crucial for today’s retailers. Let’s look at one of the key aspects of the role of a category manager at a retailer: choosing the optimal price points for all the products within the categories she is responsible for. Without automation, the category manager manually had to check out competitors’ website to determine what prices competitors are currently charging for their products. She puts this data into an Excel and makes some calculations to arrive at a decision. It is a limited data set of current pricing, so the business logic will likely be something like “adjust the price point to the lowest price of a predefined list of key competitors”. Then she changes the prices of those products manually and one-by-one in the retailer’s ERP (Enterprise Resource Planning) system. This is a very labour-intensive and — let’s be honest — an extremely boring process. That is why we typically found category managers at retailers that didn’t have automation in place updating the price points of their top selling products only once a week. They either weren’t touching the prices on the rest of their assortment at all, or they were updating them with a frequency that was significantly lower than even the already low frequency of once a week for the fast-moving products.
Now contrast this with a category manager having the support of software like Omnia Dynamic Pricing. Omnia automatically gathers all competitors’ price points on the full assortment and it can even refresh that data multiple times per day, if needed. Via a user-friendly web-based interface the category manager can automate any pricing strategy he or she can think of, on any part of their assortment. The software makes all the required calculations on huge sets of data, based on the strategies the category manager has entered into the system. Instead of simple business rules like “follow Amazon” or “set the price to price position 3 on Google Shopping”, the AI baked into Omnia gives the category manager access to intelligence that has never been possible before: dynamic pricing based on price elasticity of products. This is enabled by Omnia maintaining a history of all data points gathered and determining price elasticity of products by applying machine learning models on those huge sets of data. The system could learn, for example, that a certain Ultra-HD television is highly price elastic while the wall mount that is almost always cross-sold with the television is inelastic. That means that it will make sense to price more aggressively on the television — as that will lead to huge volume uplifts — while there is room to take more margin on the wall mount. Note that behind the apparently simple feature of pricing based on price elasticity — available at the finger tips of the category manager via the software — are very complex statistical models applied to huge data sets with huge computing power. That is orders of magnitude different from doing some calculations on a very limited dataset in Excel on their laptop.
With the software the category manager can now easily test different pricing strategies on various parts of their assortment with just a few clicks. Through this capability, they can test, come up with new ideas based on the results that are also displayed in the web-based interface, and adjust the strategies based on that. Instead of doing the boring work they did before of manually checking prices online and doing some calculations in Excel, she has now entered a creative learning loop of testing pricing strategies at scale. Because of the Marketing module that is also available in the Omnia software and that automatically calculates online marketing bids for their online marketing colleague, she is also sure that pricing data such as price ratio and price position in the market are taken into account when products are advertised. For the category manager this means a further improvement of the price perception of the retailer. For the online marketeer it means a huge improvement on the return of online marketing spend, as pricing is the number one factor impacting online marketing returns for retailers. The automation and AI baked into Omnia not only gives the category manager and their online marketing colleague superpowers, it also makes their jobs much more creative and attractive.
At Omnia Retail we feel a responsibility to make sure society benefits from AI. Both from the short-term perspective of the impact on people’s jobs now, as from the longer-term impact on society. For the short-term perspective, while AI gives the category manager in the above example superpowers and makes their job more attractive, it does require a somewhat different skillset than their “old job”. Before, the role was primarily a trade role and negotiating with suppliers was the most important element of her role. The role has now become far more data-driven and they must be comfortable in using this powerful technology. At Omnia feel a responsibility to play an active role in helping our retail clients in the training of their workforce. We do this either through personal training by Omnia Solution Consultants or by our consulting partners, or through digital means like the Omnia Blog and our (closed) e-learning environment Omnia University. It is deliberately not our mission to automate away people’s jobs, but it is our mission to Reinvent Retail by giving the retailers’ employees superpowers.
From the longer-term perspective of AI reaching a level of superintelligence that has the risk of potentially having negative consequences, I think the best thing I can do is to not reject a technology that is so powerful that it is simply unstoppable, but to actively participate in it, fully understand it, and make sure we steer it in a direction that guarantees AI is used for good. Or as Kevin Kelly has phrased it so eloquently in his TED Talk: “It’s only by embracing it, that we actually can steer it”.