Everyone is selling AI nowadays - or are they really?
The hype of AI is bigger than ever, and most technology vendors with high aspirations claim to have AI solutions of some sort. But what are they selling, really? Sogetis Digital Strategy Advisor Essy Dahlin dives into the topic in her latest blog post.
The hype of AI is bigger than ever, and most technology vendors with high aspirations claim to have AI solutions of some sort. But what are they selling, really?
To start with, let’s clarify the concept of Artificial Intelligence. AI is a vast landscape of different technologies, logic, methodologies, and algorithms, and covers multi-disciplinary fields such as mathematics, computer science, psychology, linguistics and neuroscience. But in short, the common perception of what AI is about is technology that can learn and interact with human beings. So, what does that mean? Our cognitive capability is one keystone in that concept. Learning by gather information, interpret, reason and act accordingly. Then gather information from the result of that action. And so, it goes on. We are all formed by our experiences throughout our lives and adjust our behaviors without even reflecting on it.
Translated to machines, what we are talking about is deep reinforcement learning algorithms that constantly learn from previous experiences, as Deep Mind’s Alpha Go Zero does when it plays Go with itself to improve its own game playing capabilities. Traditional machine learning is often used in the solutions on the market today, which can learn from a set of data to perform a specific task, but won’t adapt to new sets of data without the help of a data scientist. Deep reinforcement learning algorithms, however, will do that.
Drawing parallels to linguistic capabilities, an AI machine would learn and improve its understanding of intentions and syntax from each conversation it has with a human being, regardless of whether it's written or spoken.
“But remember, AI is not always necessary to meet your demands.”
For a Swede like myself, the notion of implementing cognitive text analysis in Swedish, is very compelling, since we, as every country, have vast amounts of unstructured data in our own language that needs to be interpreted, both to comply to GDPR legislation but also to gain insights and improve our customer service. Having that said, many vendors who are offering AI solutions and talks about its cognitive capabilities don’t really walk the talk. At a closer examination, their solutions are often revealed as not really AI, but regular expressions on an indexing platform.
Here are 5 questions you can ask your vendor to determine if a product is AI or not.
1. How do you categorize your product? Cognitive/machine learning/AI? Describe that category.
2. What type of machine learning algorithms does your product use and in which components of the product are they used? Supervised and/or unsupervised models?
3. How does your product work with missing or incomplete data?
4. Describe how the product improves its outcomes and how to adjust for bias and opacity in the datasets.
5. What benefits does AI bring to solve our problem compared to traditional systems?
But remember, AI is not always necessary to meet your demands. You can come a long way with regular expressions, tokenizing and indexing. But it's important to know the difference in order not to be disappointed and to set the right expectations for the organization. If you're not an expert yourself in this field, I advise you to bring a data scientist or a cognitive solution architect to the table. To get the right answers you need to ask the right questions.