Firstlight Media’s Chief Scientist, Jerimiah Hamon, has clearly got around a bit.
During a fascinating discussion on data versus gut instinct at this week’s TV of Tomorrow (TVOT) Live event, Hamon revealed the unsightly innards of some of his former employers’ dodgy data practices, while claiming that his current employer’s emotional approach to data puts it “way out in front” of the competition.
Hamon’s belief is that companies do not need to harvest personally identifiable information (PII) on users to create personalized experiences. There is an obsession with slurping up incalculable volumes of PII data, particularly from monster technology companies that have the resources to set a bar for personalization practices void of PII data. Unfortunately, these companies quite simply cannot be bothered to change.
Microsoft is one notable example from Hamon’s extensive résumé, as a computational mathematician with a background in military intelligence, healthcare and medical research. He scolded Microsoft for collecting heaps of PII while refusing to build out a system where it doesn’t use personal data – urging the industry to take a stand and make fundamental changes to the way data is collected, handled and processed.
He spoke with particular fervor about recent work on sentiment analysis, with Firstlight Media figuring out ways to measure emotions in viewers. This technique studies subjective data through a blend of natural language processing, text analysis, computational linguistics, and biometrics.
Of course, Nielsen’s old hat portable people meter came under fire as a method with poorly designed surveys that end up producing results with heavily biased outcomes. The reason this method has withstood the test of time is because computers have not been able to effectively measure emotions, although Hamon feels the industry is making progress in correlating and automating the links between stimuli and viewer responses to enable more precise targeting of video content.
“The industry has struggled for years on discussions about determining genomes for content,” he declared. “We are trying to find the parts of content that make it successful. We all have predilections that evoke something inside us.”
By breaking content down into its constituent parts and learning from aggregate large populations of people, Firstlight Media is preaching a message that you don’t need to know the identities of people. In fact, you don’t really need to know anything about them at all.
“We can understand consumer desires and give them things without intruding into their lives. It’s unnecessary,” he implored.
Hamon’s militant stance on PII data is apparently quite common among data scientists, through fear that they will one day become wrapped up in a data leak scandal, or worse – be the ones whose personal data gets exposed.
“As an economist, you learn what humans will do based on math. Amazon has a team of hundreds of economists – looking at what it is that pulls people in or drives people away from a product or advertising, whatever it is,” he added.
Firstlight Media’s unique approach to data comes with a sizable caveat though, based on the cash cow methodology that is rife in this industry – where companies want to make money hard and fast using the cheapest available technology to make it run.
“They don’t spend on maintenance. They might be running a Model A Ford and could go out and buy a brand new Ford, but they don’t,” continued Hamon, implying that Firstlight Media is on the expensive side.
Not forgetting to put on his sales cap briefly, Hamon described using best of breed methods to develop so-called long-lived systems, ideally lasting a decade, although he then admitted that is overly optimistic. “By using Kubernetes, cloud-native, and SaaS – things that make it really expansive – we can reach any country in any language, and can run massive sets of data through it in real-time in many cases,” he added.
Embracing Kubernetes allows Firstlight to break tasks down into manageable chunks, solving each simultaneously, while in the old days it was all about adding things up sequentially, without techniques such as hyperthreading, a technology invented by Intel nearly 20 years ago that allows a CPU core to work on two problems at the same time. Now with the expansion into the cloud, if you want to solve five things simultaneously, you can run five CPUs simultaneously to power massive amounts of data.
Taking a comedy show as an example, Hamon explained that all the constituent parts of this title – such as run rate, how long certain characters or objects were on screen for, and what they said on the screen – all contributes to creating emotion in our minds as we watch. Those ingredients all add up to “I loved that show” or “I hated that show” – or somewhere in between.
Firstlight Media has recently passed the year mark since salvaging the remaining Quickplay assets from AT&T, with backing from private equity firm Highview Capital. When interviewing the company’s CEO and CTO together shortly after this deal completed in March 2020, Firstlight told Faultline that its roadmap into security, data, advertising and monetization was ahead of the curve. The two execs spoke about rearchitecting existing platforms rather than targeting greenfield deployments, based on data that these platforms have accumulated over the years.
At the time, it was difficult to picture how Firstlight could possibly compete with rivals boasting algorithms that could blow Firstlight out of the water, but attacking the market from the angle of a personalized future without harvesting personal data looks a solid approach in the current data climate.