As an expert, I’m often asked: what will this year bring? I don’t have a glass ball to look into the future, or an artificial intelligence (AI)-based system for these kinds of predictions, but there are some interesting trends I certainly want to share.
I will not discuss the growth figures of AI use cases, or whether those not using AI will limp along behind, or whether the “AI bubble” will burst and a new AI winter will come. Much progress has been made, but not enough to deflect the next hurdle: and that is how to gain knowledge about your business domain with the help of AI.
So, what will AI bring us in the near future? Let me discuss three important topics:
Machine learning is here to stay
“AI is already starting to transform how organizations do business, manage their customer relationships, and stimulate the ideas and creativity that fuel ground-breaking innovation.” (Capgemini)
Three years ago, good use cases for machine learning were hard to find. Now, success stories are everywhere. Machine learning, deep learning, neural networks, and all the other variants are now plentiful. So, whatever will happen this year, machine learning is here to stay – and, it will become even more successful as more businesses start to use AI for their daily activities.
All these AI algorithms now constitute an integral part of many data-driven tools. For data analysts, using AI is just a click away. But does this imply that AI is used correctly? I’m afraid not, because:
- Data quality is still a major issue. Without data quality, bias and prejudice are just around the corner and the output of the AI will not be accurate. Data quality and ethics are intertwined.
- Ethics is often regarded as something extra, but it should be at the basis of any AI-implementation – or, for that matter, at the heart of any other big data project.
- There is still not enough focus on the quality of the outcomes of AI. Statistical methods such as accuracy, precision, recall, and F1 Defined, are definitely good indicators. But we still need a bar to measure against. Many organizations don’t really have a clue about how to measure the quality of the decisions they make.
But there’s more to business processes than task execution. How can we determine if our AI is really an improvement over human-based actions? This is still an open discussion.
Currently, we see machine learning being used in very narrow applications, to make process steps more efficient or to alleviate tedious jobs. But how AI will contribute to a meaningful return on investment has also been a big question, both last year and in 2020.
Discussions about ethics will continue
“AI ethics isn’t just a feel-good add-on — a want but not a need. AI has been called one of the great human rights challenges of the 21st century.” (Khari Johnson)
Last year, discussions about the ethics of AI really took off. Though mainly academic, the discussion now focuses not only on the (im)moral consequences of AI, for instance discrimination, job loss, inequality, and so on. The focus now is on values. Is there a thing like “AI for good”? Do we as a society really want to give decisive powers to machines? And are those machine fair and open? And what about checks and balances?
These discussions do not focus on AI alone. They also concern the use of big data. Smart cities, facial recognition, fraud detection – these are all areas where privacy and expedience are to be discussed and assessed. This will require the evaluation of the ethical side from the beginning of the project. Will the ethics of AI be a burdensome duty or a real competitive advantage? I don’t know yet.
We will see the rise of ethical frameworks. Just like compliance frameworks for accounting, these frameworks will offer ways of assessing the ethical implications of AI. Like any framework, they are no excuse not to think independently and systematically about AI. Frameworks don’t guarantee a good outcome. And the discussion will arise on how to use these frameworks in a business context.
Reaching for knowledge
“Deep learning has instead given us machines with truly impressive abilities but no intelligence. The difference is profound and lies in the absence of a model of reality.” (Judea Pearl)
Machine learning, including deep learning and neural networks, is highly successful. These methods are all very good in extracting information from data. Yes, I’m aware of the numerous mistakes machine learning makes, and about how machine learning, mostly image recognition, can be fooled. We must learn from these mistakes by improving the algorithms and learning processes. But AI of far more than machine learning alone. Cognitive Computing, Symbolic AI and Contextual Reasoning are also AI. We need to re-evaluate the use of these other AI- techniques for our applications.
This year, we’ll continue to open the black box of machine learning. The algorithms will, through interpretable machine learning, provide insights into how they reached their decisions. But AI in a business context will not be able to evaluate the correctness and fairness of the decisions.
Machine learning is good at extracting information from data, but it’s lousy at extracting knowledge from information. For data to become information, it must be contextualized, categorized, calculated, and condensed. Information is key for knowledge. Knowledge is closely linked to doing and implies know-how and understanding. This raises the decades-old philosophical question of AI: “Do AI systems really understand what they are doing?”
Without visiting John Searle’s Chinese Room again, I truly think that the next step in AI can only be taken once we incorporate some level of knowledge or understanding of AI. In order to do that, we’ll have to take another step toward human-like AI. For example, by using symbolic AI (or classical AI). This is the branch of AI research that concerns itself with attempting to explicitly represent human knowledge in a declarative form (i.e., facts and rules). Combining these older techniques with neural networks in a hybrid form, will take AI even further. This means that causation, knowledge representation, and so on are key factors necessary to take AI to the next level – a next level that will be even more exciting than the achievements AI has reached this year.
For more information on this connect with Reinoud Kaasschieter.