There is a lot of overhype when it comes to artificial intelligence (AI), with many enterprise software vendors overselling their AI and machine learning (ML) capabilities.
This was the word from Jared Molko, founder and CEO of AI-based job placement platform Tellos, giving a keynote presentation at ITWeb Artificial Intelligence 2019, held today in Johannesburg.
Molko previously spent seven years at Google, where he worked in a variety of roles across Africa, Europe and the Middle East.
In his presentation, titled: “Adapting to an AI world: The Good, The Bad and The Ugly”, Molko highlighted the importance of organisations having a “decent level of scepticism” and asking their service providers some tough questions, when embarking on their AI journey.
Many enterprise software vendors, he explained, are focused on the goal of simply building and marketing a technology product rather than identifying a specific use case and the business value to customers.
“There is a lot of overhype and overselling when it comes to AI; everyone is claiming to have some sort of AI capability when they actually don’t. I’ve visited the Web sites of vendors who claim to offer machine learning, deep learning capabilities, all these amazing services, when they really don’t. We need a healthy dose of scepticism regarding what businesses are getting fed by their service providers,” he advised.
The problem with technology as a whole is that investors and engineers are impatient; they race to the next big thing without interrogating how it’s going to transform the businesses and services, he added.
“While AI is really a great, powerful and valuable technology, there is a poorly thought philosophical approach to it. I’ve been in boardroom discussions where we try to understand certain challenges within an organisation and the term 'machine learning' often comes up as the ultimate pacifier, without any questions interrogating how exactly the problems will be resolved.”
Becoming reliant on, or handing over agency to nascent technology without fully understanding what’s going on is risky business, warned Molko.
He stressed the importance of organisations having a certain level of literacy in order to ask services providers the relevant questions, such as: What is the framework of the algorithm that is in place? What data is being used? How does the algorithm figure out its outcome? How does the algorithm work with the various components of data? How does the organisation gain overall business value?
“There have been so many situations where companies deploy AI programs and don’t fully understand the data they’re using or the algorithms they deploy. They also don’t bring in the relevant subject matter experts involved and leave moral, ethical decisions to engineers and mathematicians who are simply ill-equipped to deal with such things,” Molko pointed out.
So the onus is still very much on the organisation taking on any AI project to make sure it covers all its bases, “otherwise this beast can run away with itself and it will cause major damage,” he warned, making reference to AI biases and ethical issues.
Wrong framing around AI
There are three main philosophies around AI, and businesses often approach their AI perspectives based on these three: machines completely take over businesses; man and machine work together; or the concept of building half-man and half-machine agents, he explained.
“We have the framing all wrong and that is why there is a tremendous amount of fear and anxiety around AI. This framing is very counter-productive because it makes it appear as though AI will replace all jobs when we should bring the conversation down to its real terms and focus on the infinite opportunities it brings,” he continued.
Data is liquid gold
Data, explained Molko, is known as the new liquid gold, but it is also fast becoming the new snake oil, and a healthy level of scepticism will help businesses make the right data-related decisions.
One such decision is ensuring they hire a data scientist who has all the relevant skills – which he believes can be a near impossible trait to find.
“Nowadays, we have a new crop of people coming out of the woodworks and calling themselves data scientists. But to be a true data scientist is a very difficult thing. One of the key aspects missing from many true data scientists is domain expertise – they need to understand the business and the industry they are working in; if they don’t, it can lead to a whole host of mistakes,” he went on to say.
“Data scientists also need to be statistical experts, programming experts, visual and communications experts, database technology experts, and they need to be able to use various technologies to integrate and extract insights from data effectively.”
All these skills are tremendously difficult to have in one, and unfortunately, our eduation system doesn’t fully equip students to be fully-fledged data scientists upon leaving tertiary institutions, he noted.
“Being a good data scientist is something that’s learnt over time. Research shows that 80% of data scientists are employed by the top Silicon Valley organisations, so this means there really aren’t a lot of true data scientists out there because many people who are calling themselves data scientists are actually not,” he concluded.
Share