Yes. In the long run, we agree that artificial intelligence will deliver dramatic improvements in our jobs, lives and societies. But in the short run, the hype around AI makes for inflated expectations – and many companies today are getting disappointed that their investments in algorithms are not paying off. At the peak of the hype, an AI crisis is unavoidable.
In the first decade of Python Predictions (founded in 2006), almost all our projects were about building algorithms to predict behavior, events and attitudes in a large variety of industries and business domains. Based on what we see in the industry today, We’ve listed five bubbles we expect to burst. And we’ll conclude with how these insights pushed us towards a new AI mission.
- WTF is AI? Artificial intelligence today means very different things to different people. It’s important to realise that the container term AI captures very mature components (like figuring out who should be granted credit, predicting who will watch which movie etc), and applications that are maturing as we speak (such as computer vision, self-driving cars and chatbots). In general, there is quite some confusion about what AI really means and covers – and the technology needed to solve the problem. The danger is, that many companies today (startups as well as mature players) benefit from this confusion to simply (re-)package existing products and services as “AI-driven” merely to attract funding. Building a business based on confusion does not seem like a viable long-term business strategy, if no actual added value is created. This is the first bubble we expect to burst – since in the long run, common understanding will replace confusion.
- In it for the show. We see an increasing number of companies where managers are launching AI-related projects “because they are cool” – for the visibility of their companies or personal merit. Additionally, we see an increasing number of data scientists blinded by their love for using novel technology – only focusing on what’s new and popular now. Neither of both put value first – while even Elvis knew it’s “one for the money – two for the show”. In the long run, what is cool and novel will change – and the domain will only survive if actual value is being created. This is the second bubble we expect to burst.
- The paradox of speed. This is related to the previous bubble. As it is cool to work with novel technologies, companies renew their data and analytics architecture and infrastructure continuously. Today, there is an ever-increasing pace of technological change, and for some of the companies we meet, this means in practice that they have no idea about how the algorithms they build today will be in production next year, since they don’t know what their production environment will look like in a year time. In other words, change happens so fast that these companies are actually at a standstill – this is what we call the paradox of speed. But if the algorithms fail to be put in production, no value will be created. This is again a bubble we expect to burst.
- In the algorithm we trust. This is what amazes us on a daily basis: when we launched our business in 2006, we spent a lot of time convincing and explaining to business people that algorithms often outperformed human decision making when data was available. Today, we sometimes see a blind trust in the algorithm – where black box algorithms are accepted to make important business decisions. It is important to realise that this not only poses a risk, it also leaves a lot of value on the table, as interpretability and predictive performance both are capable of delivering business value. And there are many ways to ensure algorithms are interpretable. Fortunately, there is pressure coming from thought leaders focusing on ethics and legislation, all in favor of interpretability. We expect the blind trust in the black-box algorithm to be the fourth bubble that will burst.
- Mind the gap. Indeed, many organisations have leaders who believe in the value of algorithms. And many of those succeed in attracting a group of enthousiastic (young) data scientists. And they are on speaking terms. But more often than not, they are lacking a constructive dialogue needed to make progress. In other words, in many organisations, managers and data scientists do not speak the same language – and while the idea of deploying analytics translators is not entirely new – we don’t often see the roles emerging in reality, leaving a huge gap between management and data science teams. In this gap, a lot of value is lost. Thinking that ‘value will be created as soon as you have managers and data scientists on board’ is a fifth bubble that will burst.
And we’re to blame too. We too let ourselves be blinded by cool innovative projects in our initial years – and we too have focused on gathering the coolest applications in our domain. In fact, our company mission until end of 2018 was:
“nailing the coolest data science projects”
which meant in the first place, working on the coolest projects, but in a way clients would say “they nailed it”. While this mission was certainly fun, in today’s context, we decided to take a step back. We realised that we are not Google, we are not Facebook, we are not Amazon, Apple or Baidu. We will not invent the driverless car in our Brussels office. So we reflected with our team and core clients about what we do better than others, and packaged that into our new AI mission:
“being the most valuable AI business partner”
which basically means our core business is still being a data science team, mastering AI and the latest trends in technology. But we’ve strengthened focus on applications and technologies that are relevant in a business context, and in projects where the business understanding is of crucial importance to cracking the case. As always, we value and build partnerships with our clients and always strive for creating maximal value for the companies we work for. We’ll continue to bridge the gap between management and data scientists by strengthening our offering in workshops for managers, data scientists and yes, even analytics translators.
In most recent years, however, we’ve also directed our attention and investments towards other crucial data topics, such as data strategy, business intelligence, data foundations and information management – all very complementary to AI. After all, the grandest challenge lies not in delivering a first AI use case – it lies in building truly data driven organisations.
