You will receive a copy of your message in your e-mail.
Actualize Partners is excited to announce the launch of our newest product: Perceiver AI. Developed completely in-house by Actualize Partners, this powerful and novel AI service is an out of the box solution that reduces a company’s need for dedicated data scientists or other human intervention to solve its most complex optimization challenges.
This is accomplished through the automated analysis and understanding of difficult-to-identify data patterns, and ultimately the generation of exceptionally accurate predictions.
Sounds like a mouthful, doesn’t it? I’ll break it down.
Businesses need to make decisions based on data. That might mean an investment professional making a recommendation for a client’s stock portfolio, or an HR department sifting through thousands of resumes in order to determine which candidates will be called in for an interview.
AI Fills the Gap
But all that data — the historical performance of certain stocks and the thousands of resumes, in these examples — is too much for one person (or a team of people!) to handle. That’s why artificial intelligence has filled in that gap, using machines and machine learning to aggregate and analyze existing data, learn from new data, and then use that data to develop models that can be used to make predictions or recommendations that guide decision-making.
Despite its wide application across industries, AI is still a fairly new technology. Around 15-20 years ago, AI was mostly something you’d only find at universities and big tech companies. Most AI applications are “neural net based,” which is a fancy way of saying they try to mathematically model the human brain.
Two of the most famous examples of AI occurred in 1997, when Deep Blue defeated world chess champion Garry Kasparov, and again in 2011 when Watson beat Ken Jennings on Jeopardy! However, AI is about more than man vs. machine. Today, AI is used in fields as diverse as financial services, insurance, education, and healthcare in order to solve data problems that humans alone cannot.
Current AI Shortcomings
Still, AI is not without its shortcomings. Current AI applications take the data in and spit out a prediction or recommendation, without offering any insight as to how it arrived at that answer. This leads to a reproducibility issue and also doesn’t foster trust in how to use AI in sensitive domains such as healthcare. And since we only have neural net based AI technologies available to the general public, it’s like the old cliche that when you have a hammer, everything looks like a nail.
Furthermore, data scientists tend to be heavily involved in the model making process for AI and machine learning products — which opens the door not only to human error, but also to the introduction of biases that negatively influence decision making down the road.
And this is where Perceiver AI comes in.
Even though we are just launching commercially now, research & development for Perceiver AI began about 15 years ago. We were inspired to create Perceiver AI in order to remove the challenges with current AI products on the market. As AI was becoming more broadly adopted and interest in the technology grew, we tapped our excellent 115 person nearshore team to complete development.
Our R&D process started with our hypothesis, which was that the current AI (neural-net based) can’t be the only thing we need to address different types of problems related to working with data. Then we asked ourselves: what are some of the current AI applications today, what’s the state of knowledge about AI, and what are the data problems out there that could be solved with a new approach to AI?
Our research led us all the way back to Alan Turing, who suggested back in the 1950s that genetic programming could be applied to AI. “Genetic programming” is an evolutionary approach to machine learning, taking the Darwinian concept of “survival of the fittest” and applying it to computer programs’ ability to learn and evolve based on new data it receives.
However, attempts to use genetic programming in previous AI applications were not highly successful. So we wondered: why don’t we create an AI application based on these earlier concepts, where computer programming evolves/learns for itself according to genetic programming? We read academic papers on the subject to find what the known problems were, and then worked on solving them so genetic programming could be used in AI. And that’s where the real fun began.
The core of good R&D isn’t just the science; there’s an art to it as well. We believe that R&D is most successful when it occurs in an environment of “controlled chaos,” a strong supporting infrastructure that enables rapid testing, prototyping, and tracking results.
This combination of factors let our researchers be creative and get sparks of ingenuity to unlock the full potential of AI. Having the right culture is important too. We empower people to take educated and thoughtful risks that can have high payoff.
Our researchers have high standards and accountability, but also recognize that it’s okay to fail as long as they learn from mistakes and strive for continuous improvement. When the science and the creativity align, that’s where the magic really happens.
We developed Perceiver AI on the side, not as our full-time focus, until a few years ago. During that time, we tested Perceiver on multiple use cases and refined it as we gained more insight into how it works in the real world. This process of constant iteration led to the product we have today.
A Proven Breakthrough
The genetic programming approach taken by Perceiver AI relies solely on observed data patterns, which means unlike other machine learning tools, it is not subject to human biases. For businesses that already have solutions they want to improve upon, Perceiver can use existing algorithms as a starting point so it’s not beginning from scratch, which saves valuable time. Perceiver is also designed to be highly scalable, since it is able to accommodate expansive datasets using distributed programming principles.
So, what can a tool like Perceiver AI do for a real life organization? Well, it has a wide range of applications across industries, ranging from logistics to financial services to healthcare. In an initial case study, Perceiver AI analyzed data from a commercial airline’s fuel usage and developed a new fuel tankering model that resulted in 30% incremental savings for flights departing from one of its main hubs.
Perceiver AI can also be used to track and predict investment performance. We have been using it in this capacity for the past three years with exceptional results, achieving well over 200% cumulative returns.
What’s Next
To learn more, click on any of the links in this article to visit the Perceiver AI website.
Contact us to discuss AI projects/engagements with Perceiver and also any other currently available commercial tools such as TensorFlow, Watson, and PyTorch.
As a side note, Perceiver AI truly demonstrates the strong, complex engineering capabilities of Actualize Partners. And it’s also representative of the R&D process we use not only in-house, but with our clients.