Artificial Intelligence vs Real Stupidity: Busting 5 mythsvendredi 12 janvier 2018
AI (Artificial Intelligence) is the buzzword at the forefront of technology nowadays. You can’t spend more than 5 minutes browsing business or tech news without some bond-villain-esque billionaire telling us we are in the “golden age of AI”.
In today's blog, we’re talking about the AI hype, more specifically the AI over-hype that we are seeing everywhere - from the news, to magazines, to airport bookshops, to crazy and delusional reports from other consultants, to companies bragging of their AI chatbot.
People, and more importantly businesses, are forgetting that AI and analytics run on the same fuel, both require data, and there is no point ploughing your hard earned bucks into AI until you’ve mastered your data and analytics, which is even more important in the face of a rapidly brewing GDPR storm. A company blindly running into AI because they have read about it in Forbes when their idea of analytics is still Excel and a bar chart, is more an example of real stupidity than it is artificial intelligence.
The rate of innovation and understanding on the research side is increasing rapidly. Achievements like Google DeepMind’s AlphaGo Zero computer program - which achieved a standard beyond human capability at a game of Go; while also being entirely self-taught - are major milestones on the road to creating general purpose intelligent algorithms.
However, as much as research achievements are hugely exciting, this has led to over-hype on the corporate side. VC investment into AI in 2016 was over $3.6bn, with the likes of Baidu and Google spending closer to $30bn between them. That’s all well and good when you’re one of the biggest tech companies in the world but for the smaller firms that lack the data and tech stack, it is borderline insanity.
We can’t stress this enough - until you have mastered and embraced analytics you are not ready to step up to AI. Usain Bolt is the fastest runner in the world, the best at what he does, but put six month-old Usain on the starting blocks in Rio and he’s going nowhere. You can’t run before you can walk.
The over-hype of AI has resulted in a number of myths that we hear over and over from companies who are just finding their feet in the world of AI. We address some of those myths here…
Myth 1: AI and Machine Learning are the same thing
People often get AI and machine learning mixed up but is it really the same thing? Our way of thinking about AI is by analogy with human intelligence. A human can collect data via the senses, for example by looking at another person. Our brain can then process that data, resolving signals from the rods and cones into an image of the person. We then have an understanding of that data. For example, we might understand, using prior experience, that the person is smiling and offering their hand and that they expect us to shake their hand. Finally, we can take an action, by shaking their hand.
In a similar way, for an AI system, we expect the same four capabilities: data collection, data processing, machine (in the artificial case) learning, and action. This is a more advanced version of the conventional analytics structure of data collection, data processing, data analytics, and action. While the AI system encompasses the full ecosystem from data collection to action, machine learning is the instrumental cog in making it all work – it is the brain of the machine.
Myth 2: You need AI to win
HBR found that companies in industries furthest along the digitisation curve, such as telecom and automotive, were 50% more likely to see profits from an investment in AI than less digitised industries. Look, for example, at Netflix, who are already famous for using data and analytics to craft compelling content – are now putting AI to work on all their data and are reaping the rewards. Netflix claims that their AI-backed recommendation engine has helped customers quickly find content they want to watch, reducing the number of cancelled subscriptions and saving them $1bn in lost revenue per year, almost enough to make 77 episodes of The Crown.
It’s also important to note that AI is far from a sure thing. IBM is one of the smartest companies in the world according to the MIT Technology Review, and who could claim otherwise? But even IBM overhyped their own AI. In 2012, Watson – IBM’s question answering computer system capable of answering questions posed in natural language - started working with the MD Anderson Cancer Centre. In 2013, IBM claimed Watson could carry out clinical trials and in 2015 Watson was said to be learning to do what doctors can’t. Then in 2017 IBM and MD Anderson parted ways, with a $39m bill to MD Anderson and nothing ready to be used outside of pilot tests and , according to the Wall Street Journal.
This is the issue with AI - it’s a fight to be on edge of development but companies are running blind. Tech Pro Research found that 50% of companies they looked at want to start using AI soon, while only 28% of managers in IT leadership positions had any experience with AI. AI is risky and expensive to develop so you should think twice before splashing the cash in the race to innovate.
Myth 3: You need Machine Learning
Ask yourself: is AI a good fit for my business? We often see businesses, especially in the B2B space, that do not want to close the loop from data collection to action: for these clients the role of the data is to provide insight, but the decision-making and action will remain with the management. In this case, the fourth AI step of machine-operated action is not required, while machine learning or advanced analytics platform - sitting on a high quality data platform - is sufficient to allow the business to succeed.
However, it should still be considered whether machine learning is the right tool for your business. We often see businesses where the insights they need could be drawn from the data by conventional analytics, but where they are not yet using it to its full potential. In this case, the answer is not AI or machine learning, but better data management, powering analytical tools. For many of our clients, this solution is more than enough and is quicker and easier to implement and troubleshoot than machine learning. If you can do it with a regression, do you really need that machine learning algorithm?
Myth 4: Robots = AI
When most of us hear the letters AI we think of Siri or Cortana in our phones or those silly looking Google cars. Few of us think of the dark Minority Report side to AI and the repeated warnings from Elon Musk about how we should be scared and cautious of AI robots (although that is a discussion for another day). A few of us might even think of the latest iRobot Roomba 980 but not all robots have to possess AI.
In situations, were the data input is unlikely to change, it is almost always more feasible to work with robotic process automation allowing a robot to act on data through pre-specified reaction programming than to go through the full suite of developing AI. Think of the Ford Motor Company that managed to reduce assembly time for a single vehicle from 12 hours to about 90 minutes by breaking the assembly line into 84 distinct steps. As a result of this highly compartmentalised assembly process Ford later on was able to significantly reduce production costs by introducing robots without a shred of AI.
Myth 5: All you need is data
What brings together data, analytics, tooling, optimization and yes ultimately AI, is your analytics strategy, not just what Tableau dashboard to use, but where in your organization analytics can be most transformational. Laying the bedrock of your analytics strategy across your company will set you up to capture the benefits of advances in technology as it happens, instead of trying to bolt something on to your old school ERP system when it becomes trendy.
Just as on a foggy day (one of the co-authors is writing this at Mumbai airport where the weather forecast is officially 32C and smoke!) the pilot is challenged land your plane, a decision based on visual information isn't always possible. In the same way, your business cannot make intelligent decisions if it does not collect high quality data. Just as if I cannot resolve the colours and shapes into images of objects I understand, your business cannot make intelligent decisions if it does not process its data well into a format that is reliable and can be easily understood.
The AI hype has grown out of proportion. While various sources may say that retailers are already benefiting from AI-powered robots to run warehouses, the fact of the matter is that most of companies are still nowhere near being able to extract value from AI. Those who attempt it are playing a high risk, (probably zero) reward game. Instead, most are and should be focussing on fixing their foundation for data collection and processing which will allow them to realise substantial value from conventional analytics which has yet to be unlocked.
Thinking about AI is great but it cannot be implemented before you have the basics in place and surely, a machine making decisions on bad data will produce nothing more than artificial stupidity.
- James Walker, Partner and Global Head of Analytics
- Frederik Oliver, Associate Consultant
- John Clark-Maxwell, Associate Consultant