If it’s written in Python, it’s machine learning. If it’s written in PowerPoint, it’s probably AI
“I work in AI,” says Christopher Conroy, data consultant. He would much rather say that he worked in machine learning. His passion is putting data science models into production, but he is not happy with the description AI. “Some of the smartest people I know, in terms of building code and writing algorithms, and solving data science, have had challenges getting a role because their CV said machine learning, instead of AI.”
He cites Mat Velloso, an adviser to Satya Nadella at Microsoft, who once wrote: “If it’s written in Python, it’s machine learning. If it’s written in PowerPoint, it’s probably AI.”
He doesn’t want to pull the H word from his lexicon, “pejorative” is how he describes it.
“The Significant media focus on the acronym AI has been unfortunate,” he says, skilfully managing to avoid dropping the H bomb, namely ‘hype’. “For many organisations, nothing has changed from two years ago, they could have said then ‘we are using AI’.
He likens the current media interest in AI with the hype (whoops, said it) in data science during the middle of this decade.
“AI is a correct term but is often a misleading term. People think something has changed, but in fact many organisations probably had a team working on things like random forest models, and called the work machine learning, some time ago.”
A deep look into artificial intelligence, machine learning and data science
Moving data science models into production
Conroy has been around for some time — some time for working in this field. He refers to the year 2011, as far as data science is concerned, that might as well have been the time of Ancient Egypt’s pharaohs. He worked at Rank Group for seven and a half years, initially in BI, then set up the data science function at the company and says that by the time he left there had been “significant digital transformation” and the company had ”end to end cloud infrastructure.”
His focus, however, is on putting data science models into production.
“There might be some research team in a lab in Cambridge or Berkeley trying to solve a problem relating to the mathematics of a blackhole, but the vast majority of commercial data science is not focused on such problems. It is at the simpler edge of data science and a lot of the value comes from the most simplistic models”
However, “getting data science models into production is extremely challenging.”
What does he mean by production?
“Data scientists build models in python, maybe in a cloud environment, more likely on a laptop using sample data. When these are in production, for new customers that arrive on the website I can, for example calculate future value, and from that create a CRM journey, showing them different content on the site, and recommend what product to show each customer as it arrives. So production is where we are scoring against the customer and giving different outcomes, so it is about getting to that tangible outcome.”
There is a problem, however. Conroy puts it down to immaturity — the data science role not being around long enough. And that takes us to why organisations are not getting the most from data science.
Automating data science and machine learning for business insights
The corporate mindset
“A lot of organisations are not fully committed to data science; they have hired a data scientist and then expect them to start helping boost revenue.”
But things are not that easy.
“You don’t hire a CRM team and just give them photoshop, you give them a tool like Adobe Campaign or SilverPop, and deliver a customer journey by text or email, etcetera.
Christopher Conroy: You have some really smart people delivering outcomes, but it’s akin to a marketeer sending out marketing emails from their Outlook account. What we really need is for it to be scalable, repeatable, understood and trackable.
“A lot of people hire data scientists and don’t give them a method of production” So in part, they just hire data scientists without a thought through reason why.
“Everyone accepts, these days, that you need an enterprise level marketing tool, one enterprise tool to deal with your customer communication, and secondary marketing tools to deal with your on-site stuff. Typically, with data, the only enterprise tool is delivering dashboards.”
The future of data scientists, automation and the citizen data scientist
Data scientists are in demand, their remuneration is matching this, but what about their future, can their roles be automated? Will technology tools remove the nerd from data scientist, and open the field to a wider audience? And what about the citizen data scientist? Read here
Looking beneath the facade
Even the really smart digital companies are not immune. “There are some really smart digital companies, of whom you assume they must be good at data science. A lot of them are, but when you peek under the covers, all they have is a bespoke solution for each individual model, so no enterprise wide solution.
“You have some really smart people delivering outcomes, but it’s akin to a marketeer sending out marketing emails from their Outlook account.
“What we really need is for it to be scalable, repeatable, understood and trackable.”
Doing it right
Right now, Conroy is working with a start-up. It has been “One of the most enjoyable things I have ever done.”
Startups, he says, as a general rule, have always had one big advantage. “This is why so many start-ups are successful, there’s less legacy, they are faster and ultimately more efficient.
“The same can applies to data start-ups, they can have those same major advantages.”