Chara AI is a Scotland based Artificial Intelligence company founded in 2019. They focus on interpreting stories and ideas when in development, to help creative businesses meet the commercial demands of video streaming platforms.
Chara AI were selected as one of our Round 4 Resident Entrepreneurs receiving £12,000 of funding plus support to research their first product ‘Kototama’. We asked founder Andrew Green to tell us more about Chara AI’s Resident Entrepreneur project and their ambitions for the future.
I’m a film producer, film editor, and sometimes film financier. I’ve built international film distribution technology companies and, in 2019, we made this new company, Chara AI, to discover and develop new stories with a focus on new writers. We’re creating a system called ‘Kototama’, which looks something like Google docs with a ‘scoring’ mechanism that helps writers get paid to meet the commercial demands of video streaming platform audiences.
I used to travel a crazy amount doing film rights licensing deals in familiar locations like Los Angeles, Tokyo and London, but inspirational were deals I did in non-mainstream places like Manilla, Lagos and Chengdu. I learned a lot about people because everyone would pitch me their business and their story. When I decided to stop traveling (not good for family life), something was missing, I had thrived off the diversity of people I met and this was now gone.
Doing deals in countries that are not wealthy presents a lot of challenges. I’d learned how we (in the West) exploit people by presenting expensive legal and banking processes that must be met to provide ‘legitimacy’ to transactions.
For people out with our system, the literal costs are far higher because of the objective risk we capitalise on through excessive fees. This is unjust and in reflection, this injustice is replicated in my home country. My motivation is to make a company that is fair to all. This means all people have the right to tell their stories and be paid without discrimination.
Back in 2019, my mind was sprawling with messy half-ideas on how I’d attack the complexity of the challenges I’d set out. My friend Zahid pointed me to MIT Professor Lex Fridman, who waxes lyrical in his podcast synonymously titled ‘Artificial Intelligence’, his guests are geniuses and agents of provocation. I listened to 100 of the interviews and then read every AI paper I could find.
To get an idea of the overwhelming complexity available, read the MONTRÉAL.AI cheat sheet compiled by Vincent Boucher; here, you’ll find a discussion on ethics.
Privilege is the curse of AI because ‘doing AI’ is an expensive business and privileged people are not the most brilliant people to formulate the questions and answers required. Our challenge is to encode a system that represents very clever writers who are privileged and not privileged, and most of the time, you won’t find the latter on a university campus because they’re slaving away to deliver your order, clean up after someone else or having their screenplay hacked-up by… (you fill in the blank!).
“It takes a village to raise an AI that’s ethical, robust, and trustworthy.” — Gary Marcus, Professor at the Department of Psychology at New York University, founder and CEO of Geometric Intelligence
See the problem? It takes a village, not a king, to make an AI that works. If you’re not focussed on this, failure awaits. We all want to watch better movies yet the movie business frequently organises itself to make the movies privileged people want to make but not everyone wants to watch.
In the 1990s independent film distributors discovered Cinema-goers, in the disgracefully titled ‘fly-over’ states, actually enjoy arty-European flicks as well as the slick tent-pole blockbusters printed by the Hollywood machine.
This raises a question: if audiences want different and new stories, why aren’t we making more of what they want? Of course, the big brands are improving now, and I commend that something has indeed changed. But, we’re still measurably far from getting the stories we all want.
‘When black women are free everyone else has to be as a matter of necessity’ – Siera Dissmore – Data & Society
Privilege buys you time and forgives mistakes – your success is catalysed by access to money and networks. Succinctly the 2020 Annenberg School of Journalism paper The Ticket to Inclusion examines the ‘bankability’ of movies made by underrepresented groups.
Being a white, wealthy, middle-class man (like me) does not secure the ‘bankability’ of a movie any more than say, a black woman from Los Angeles. Instead, it is my access to money and ‘privileged networks’ that exclude people that are different from me, that facilitate an advantage in my favour, regardless of my talent, experience, work-rate, or ability.
‘To sleep: perchance to dream: ay, there’s the rub; For in that sleep of death what dreams may come…’. Hamlet, William Shakespeare
The impediment to our challenge is: how do you teach an AI on data (historical movies) when this data encodes in the bias we want to get rid of? The answer is: encode-in the missing data.
Recognition of inherited bias acts only as a substrate where we can seed intersectionality. Or as Professor Patricia Hill Collins describes, “The matrix of domination is a concept that draws attention to the inherent complexity of privilege as it operates in social systems and shapes people’s lives.”
To become personally responsible for change informs the approach we take at Chara AI. I can’t simply use all these skills I’ve got and all this privilege anointed to me to model a technology that I think is best for everyone. The entire system must be taken apart and re-shaped by all the people that access it, and this system must be highly accessible to all.
Our Resident Entrepreneur project was a 6-month mission to encode the existing diversity of our industry and compare this to audiences. Or, deconstruct the current system and then start adding what’s missing. Not ‘audience demand’ but the storytelling heritage of an underrepresented audience, or ‘the majority of people’, as it is the majority of all people missing that are from our collective filmmaking narrative.
Over the project period, data processing involved a lot of manual input because most datasets needed simply do not exist. Some of the processes applied research into the identity of Directors, Writers, and Producers.
Frankly, I found myself yelling at the screen occasionally at many talentless white men posing in woolen Newsboy Caps. Even worse, to compare their minimal ROI over their successful black woman counterparts was shocking (to this white man).
Our resulting dataset defines what to encode and presents several algorithms in how we source specific data and automated datasets to provide signals that can be accessed by downstream processes. The outcomes demonstrate a couple of new challenges. First, primarily automating ‘Consensus.’
We decided to add a blockchain part because we can’t claim that Chara AI is non-biased; we have to prove that all decisions the AI makes are agreed upon in advance and that we can’t influence this system for whatever reason. So, we successfully applied to the Innovate UK Edge programme to develop a ‘Consensus Protocol.’
First, we drafted questions based on outcomes from the Resident Entrepreneur project that humans can understand and then met with the video on demand platforms and groups representing potential users (writers) to validate these questions. Essential is an encoded consensus on what ‘audience demand’ means, as defined by non-deterministic data sources.
The users’ work is not entirely encoded into the on-chain ‘Kototama’ because blockchains can only contain compressed deterministic data. In fact, the user’s work represented by the Kototama sits off-chain, and the Kototama sits on-chain where the relationship between the two is encoded, decoded, and encrypted.
The extrinsic data is a live snapshot of ‘audiences’ but, to ensure this demand is not based on ‘historical data’ (which can’t be trusted), we need to capture non-deterministic extrinsic data and compute this within the AI, where the AI determines the validity of an update to the Kototama and offers potential state changes the AI can use to help the user.
As challenging as this has been, fortunately, we were successfully accepted to Edinburgh University’s AI Accelerator. This has provided valuable support for an investment strategy. Also, we made a successful application to Scottish Enterprise for a Smart Grant which we’ve match funded with equity investment. This support and equity will enable the completion of the Kototama app interface and build out the AI engine to a ‘commercial and accessible’ level and tie it all together in an Agent-Based Model.
There are a lot of challenges ahead, primarily the cost of HR. Finding and hiring people to work on an unprecedented application takes a lot of focus, but if you’re interested in joining our grand challenge, I offer the sage inspiration of a person born with little privilege:
‘Life is either a daring adventure or nothing at all’ – Hellen Keller.
Applications for our next round of Resident Entrepreneurs are open now and close on Wednesday 20th April at 5pm. If you are interested in applying, join us for one of our online RE Discovery Workshops, taking place on Tuesday 22nd March at 10am and Tuesday 5th April at 2pm.