Tenyks, a spin-out from the University of Cambridge, hopes to provide the next breakthrough in AI after raising $3.4 million in its seed round. Tenyks’ platform helps engineers build better AI, faster. The funding round was co-led by Speedinvest and firstminute capital and was joined by LAUNCHub Ventures, Y Combinator, the University of Cambridge, Creators Funds, Remus Capital, CSVE Ventures, RKKVC, Black Mountain Ventures, and a dozen angels.
Addressing AI’s sometimes poor results
Artificial Intelligence is seen as the future of tech, but has a mixed reputation. Bad AI is not just a staple of science fiction — where would the Terminator franchise be without Skynet? — but has hit headlines as weaknesses in AIs have come to light.
The most frequent problem has been in the training. Machine learning is incredibly good at spotting patterns that might escape human attention, but it lacks human judgement and can draw the wrong conclusions from the data it analyses. The result has been stories like IBM’s AI developed to treat cancer, that suggested treatment that would kill patients, or the AI-judged beauty contest the eliminated all the contestants with a dark skin.
Tenyks was born from the doctoral research of co-founder and CEO Botty Dimanov, with co-founders Dmitry Kazhdan and Maleakhi Wijaya added further research. The project has now come out of stealth with some impressive results to back up the academic research foundations. Speedinvest partner Rick Hao commented on the importance of this. “AI development is changing from a model-centric to a data-centric approach,” he said. “Tenyks has developed a data-centric platform that can help machine learning engineers to have more granular insights on data, assessing the reliability across different relevant data sets and then finally improve the model and results.”
Not just more AI
Tenyks are not just adding another layer of AI. Instead, their platform helps engineers improve the quality of their AIs. AI training relies on using enormous amounts of data. This makes it incredibly difficult for engineers to analyse the results, especially when the subject is outside their expertise.
“AI algorithms are programmed with data. However, this process is broken,” Dimanov told TFN. “AI engineers just throw a bunch of data in, and hope that things will work out. But this is a very one-sided interaction, we don’t have the ability to ask questions, or the interface to make amendments.” In essence, when an AI provides an answer, it’s close to impossible to verify if it’s for the right reasons.
Instead, Tenyks are building tools that use statistical techniques to distil the information the AI has learned. This, in turn, empowers the user to understand the results, and fine tune them. The team have called the innovation a ‘doctor for AI,’ and compared it to the step-changes that graphical user interfaces or search engines brought to computing and the internet, opening them up to the ordinary user.
“It allows domain experts, like doctors, to have a conversation with the algorithm,” Dimanov explained. “Then they can determine whether the behaviour is aligned with their expectations, or if it has learned to rely on biases.” A real-world example of this might be training based on x-rays, where an algorithm might learn from other data, such as metadata on a scan, rather than clinical markings, which an engineer would be unlikely to identify.
Bringing joy to AI
Tenyks’ ambition is to widen the application of AI to ordinary users. “Engineers, domain experts, auditors, and managers will have an intuitive way to understand the operation of an AI algorithm, verify its technical requirements, and amend its behaviour when necessary,” says Dimanov.
That vision even extends to school children, says Kazhdan, who is also Tenyks’ CTO. “In the same way as coding is now a required subject in schools, A.I. programming will be a commonplace skill empowered by Tenyks products.”
The funding will be used to double the startup’s engineering team, and they are also keen to ensure the company’s culture remains. “What is different about Tenyks is that we are not just about achieving. We are about happily achieving,” says founding engineer Wijaya. “We find joy in everything we do. We even measure how many times we laughed during the week.”
It is an attribute identified by firstminute capital principal Camilla Mozzolini, “The Tenyks team have a highly complementary mix of hustle, charisma and ambition, combined with serious technical capabilities.” It’s a mix that promises to transform the way the world works with AI.