A technology that brings AI and people closer – with PrecisionLoop CEO – Maciej Wolski

Can you share your background and experience in the field of artificial intelligence and machine learning? What inspired you to start your journey in this industry?

I work in the Software Development industry since 2007, but I was involved in AI research since 2014 – when I noticed that I could utilize my knowledge from the Neuroscience field at my work.

I was always fascinated by the brain’s operation and how we could create machines that could learn on their own, with our minimal help.

I consider myself a futurist, and while I spent countless hours on the subject of Machine Learning – reading scientific papers and building my own AI tools – I recognized we are still in the Stone Age of Artificial Intelligence. And the future will look different. Much different.

My professional career allowed me to understand how to build energy-efficient ML solutions (while being employed in large multinational corporations like Samsung and Demant), as for many years, I was involved in creating AI solutions for smartphones, IoT devices, and wearables.

By engaging in low-level AI research and developing a DL framework myself – I gradually began to understand that there is much more to be done in the AI sphere.

And the future is not about building increasingly large models but squeezing as much value as possible from the available resources.

This is very similar to what we see in the biological brains. And while I don’t think we will copy biological solutions – there is a lot of inspiration coming from these areas.

AGICortex, your previous startup, gained attention for its work in AI solutions. Could you tell us about its journey and the factors that led? you to become a co-founder of PrecisionLoop?

That is true. From the earliest stages, we gained a lot of interest from investors worldwide. We declared that we can build machines with autonomous learning capabilities – and while it is true, our journey ended too early. With insufficient funding – we were not able to realize most of our clever ideas.

In engineering, it is quite easy to build a demonstration in specific conditions. It is much harder to realize solutions that will be reliable in the long term. That is why we had to focus on specific areas. We chose outdoor robotics and developed a decent software package for their autonomous navigation in cities.

We also developed multiple versions of our ML framework to make our research activities easier and far more productive.

We believed that in the AI era, we will be able to get funding from investors to continue our efforts. Unfortunately, AGICortex’s Seed round was planned for the worst period in the Venture Capital market in the last couple of years.

In the end, with just 0.3M EUR in pre-seed funding – we were not able to reach the sales targets required by the investors. We were really close to raising the next round – but we ultimately failed.

All people learn from their mistakes or at least suboptimal decisions. And sometimes, the path is not about going directly to your end goal but moving in sequential steps.

PrecisionLoop’s mission is much more adjusted to what customers want to buy and what investors want to invest in. Our previous efforts in AGICortex were really ambitious – but they were started too early.

We were just too confident that investors will support our journey. And while we talked to some of the best and most active investors in the world – we did not raise the needed capital.

So at PrecisionLoop, we aim to be much more independent from the VC capital, but of course, we want to consider funding opportunities to grow faster.

What sets PrecisionLoop apart from other AI companies in the market? What unique solutions or approaches do you offer to clients?

We aim to establish a learning loop between people and AI. To do that – you need to allow even non-technical people to interact with the system in a convenient way.

The models should be interpretable, and the ways how they make decisions – transparent to both users and the system itself.

There are well-known approaches to making ML models interpretable, but we believe it is insufficient. The next step is to provide the option to review how the decisions are made and allow for making fast and precise updates.

So this is the main area where we differ from other companies and their approaches.

We allow to correct the decisions through a convenient user interface and trigger automated model re-training based on these decisions.

There are solutions for the visual interpretation of ML models’ performance – but they are targeted at IT professionals. There are no-code platforms for designing ML pipelines – but they offer something very different from us.

We at PrecisionLoop believe that people should do what they are best at. The model architectures should be designed by industry professionals. The training should be done under their supervision.

But the model maintenance can strongly benefit from the participation of the domain experts. We believe that such an approach – brings results faster and at a lower cost.

Currently, we focus on Computer Vision, mainly object detection – to deliver a proof-of-concept solution. But our product roadmap defines Generative AI (both image and text) and NLP as some of the next steps.

PrecisionLoop focuses on creating solutions related to artificial intelligence and machine learning. Can you elaborate on the specific areas or industries where your solutions have the greatest impact?

As Machine Learning increasingly influences our lives – we believe that it is necessary to democratize access to it. Farming, construction, security, and healthcare – are industries where we could strongly benefit as a society from having more people involved.

What we aim to do – we want ML and domain experts to focus on what they do best.

It is much more efficient when a farming expert works with the details of a plant disease detector than a technical person who has to figure out how to fine-tune the ML model and its hyperparameters to get improved results.

Models can be re-trained automatically based on feedback from the domain expert. And ML team can focus most on the initial phase – where the architecture of a specific model is chosen, and initial training or fine-tuning is performed.

With the stronger presence of text-based solutions, we also believe that it will be worth focusing on industries that rely heavily on processing information.

The industry regulations will force companies using AI to be able to explain how their solutions make decisions. We believe that the optimal way to do that is through natural language.

How does PrecisionLoop plan to leverage AI and machine learning technologies to address complex challenges in the industry? Could you share some examples of your upcoming implementations?

We aim to equip people with the ability to interpret the behavior of their ML models. To offer such a capability – you need to either design model architectures that support it or rely on surrogate models that distill these decision-making processes.

For many years Deep Learning solutions were described as black boxes. Their activity can be completely opaque to the user without additional post-processing.

It starts with the distributed data representations in the associative memory of neural networks. The network uses all layers and neurons to deliver the output. It is a form of compression that allows the model to contain as much information as possible. The drawback is that it is hard to interpret the model directly.

It is not like with brain scans – where you can clearly point to the active area of the brain – and define what has happened. The brain is composed of functional units divided into modules.

There is an approach in Machine Learning to mimic that structural composition – with sparse activations or isolated modules within the network.

The current understanding is that those semi-structured models perform less than standard architectures.

We want to overcome this with our unique ideas about building ML models.

AI and machine learning technologies are advancing at a rapid pace. How does PrecisionLoop stay ahead of the curve and ensure its solutions remain cutting-edge and relevant in the face of evolving technologies and market demands?

Our team is focused on cutting-edge technology and low-level AI research for many years. I personally read hundreds of scientific papers on the subject of Machine Learning each year.

We can predict trends in Machine Learning and see where it all goes.

We believe that the future is about efficiency – and squeezing the maximum value from the available resources. The benefits of AI should not depend on having millions of dollars for a single training of the largest model architectures.

The robustness of the models, transparency of decision-making processes, reliability, and cost-effectiveness – these factors will contribute to expanding the impact of Artificial Intelligence on our society.

And these are the areas our company will be active in.

As PrecisionLoop begins its journey, what are your primary goals and objectives for the near future? Are there any specific milestones or targets you aim to achieve?

First of all, we want to demonstrate the idea through a proof-of-concept solution. It will clearly present the benefits of our technology to customers and investors.

In the next stages, we want to expand the set of available features. We will support more tasks and model architectures.

Initially, we will focus on Computer Vision, but we plan to support Natural Language Processing and Generative AI (both visual and text).

We will consider cooperation with good investors, who understand the potential of this technology and will be an added value to our team.

But we are not dependent on their funding – we will work with customers from the day one.

In your opinion, what are the most significant challenges or barriers that startups in the AI industry currently face, and how do you plan to overcome them with PrecisionLoop?

Because our team is located in the CEE region – for me, the biggest barrier is the lack of sufficient funding and VC investors’ understanding of how to build such a company. I believe that companies in the USA have a much easier task.

In engineering, demonstrating the technology is just a first step. The majority of the effort is to build solutions that will be reliable - regardless of the conditions of the system. And that requires sufficient funding.

Requiring the startup to have significant sales results after the first investments – is asking to sell immature technology. I am not talking about first pilots or paid cooperation. I remind myself about investors that expect that a company that builds advanced technology will have annual revenue equal to the funding it received from investors.

In such a case, the company does not require an investor anymore.

So, in the case of PrecisionLoop, we want to work with only the best investors – who know how to scale deep tech startups.

We already have plans and strategies in place that make us independent of Venture Capital.

But of course, a good VC investor can be an added value – who accelerates growth and supports with experience and the power of its network.

Finally, considering the readership of our portal, is there any message or advice you would like to share with aspiring entrepreneurs in the AI and machine learning space?

Entrepreneurship is hard. It requires dedication. But it is also very rewarding. It is just not for everyone. If you feel that you have an idea about what you would like to do in the next 10 years of your life – pursue that goal.

If you will fail, get up and try again. Stop when you will achieve your goals. Don’t quit.

You can’t lose – if you will not quit until you will reach the place you always wanted to be in.

The recipe is simple – persistence is the key. Nothing valuable in life is easy.

#AI #ML #MachineLearning #ArtificialIntelligence #DataScience #BigData #DeepLearning #PrecisionLoop ?ukasz Dudek