Softacom has been working with AI-powered solutions for years. And recently, we’ve embarked on a new journey. Our clients can take advantage of our AI transformation services focused on neural networks and machine learning for computer vision and video processing.
In this article, Softacom’s CEO and Founder, Serge Pilko, explores the real impact of GenAI, the limitations of today’s technology, and the breakthroughs that are required to unlock the next era of AI-powered solutions.
AI is a Growing Priority
We live in curious times when it comes to technological advancements, though I suspect that every generation over the past two centuries has thought the same. The difference lies in the scale of change and the areas it impacts.
According to our clients, solutions based on generative AI (GenAI) have become a top priority for innovation in their products. Today, budgets are being allocated specifically to develop solutions with GenAI. The topic of GenAI is widely discussed, and for this reason, many companies implement it into their IT solutions and products. But it’s not just hype – GenAI is indeed a technological breakthrough in recent years. A striking example is how people now search for information: instead of the traditional “Googling,” more and more users are turning to ChatGPT for answers.
There is a clear reason for that: receiving a well-structured answer is far more convenient than analyzing Google search results on your own, filtering out noise, advertising, and SEO-optimized articles. It significantly reduces the time spent searching for information.
Many skeptics can argue that we shouldn’t blindly trust GenAI-generated responses – and they are right. But we also can’t trust websites appearing on Google or Edge. Manipulating search rankings through SEO is far more widespread than influencing the training of generative AI models, which rely on digesting information from the web and other open sources. While the manipulation of AI training data is still underdeveloped, the progress doesn’t stop. Where there is demand, the supply will follow.
I also believe that it’s only a matter of time before OpenAI and other GenAI providers introduce advertisements or something similar to contextual ads, like AdWords, into their products and search results.
If reptilians or other alien species watching us allow it, the development of quantum computing will spark a new technological revolution. The mention of reptilians is, of course, a joke (thanks to science-fiction writers), but the part about a technological revolution – that is true, I firmly believe.
The Key Challenge: Overcoming Current Technological Limits
What do we still lack to create a device, robot, or program that will evolve on its own, make decisions, and have empathy? We don’t have more powerful machine-learning models that will truly understand our world through computer vision neural networks, process what they see within the right generative model context, and give the results in an abstract form, such as text, voice, moving or taking something, or go somewhere – as a human being does.
To develop more powerful models (both for computer vision and generating AI), we need a fundamentally different computing architecture. It must consume millions of times less energy and deliver performance millions of times greater than what we have today. This shift will happen, especially as Moore’s Law, once applicable to microprocessors, now extends to AI models.
Of course, GenAI is not true thinking. It is a mathematical algorithm that can’t think – it simply doesn’t have that functionality. For those interested, I recommend watching this lecture from the University of Stanford: https://www.youtube.com/watch?v=9vM4p9NN0Ts&ab_channel=StanfordOnline. It is a technical lecture describing how generative models work. You can also find a lot of simpler explanations online, or, of course, ask ChatGPT. Just don’t forget to say hello and thank it afterward.
The more we work with AI at Softacom, the more I question whether thinking or consciousness, as humans define it, is even necessary for creating GAI (Generative Artificial Intelligence). A system could be built on a reflex-based model – reacting to its environment through computer vision models. Then, the data is passed to the GenAI models for processing within a given context (such as past observations or a, remembered plan for a month). These processed responses would, in turn, change the current context, drive actions, and so on. Why wouldn’t this be considered a form of thinking?
GenAI doesn’t have principles and emotions. But why can’t emotions and principles be called learning patterns that we, humans, also develop through upbringing, environment, culture, or life experience? School is a good example of a place where skills and behavioral patterns are instilled in us. And family? The principle is the same. Young children don’t understand concepts like respect or virtue – I often recall the book Lord of the Flies in this context (https://en.wikipedia.org/wiki/Lord_of_the_Flies).
So what is the future of GenAI and computer vision? What are we – those working in AI, neural networks, and technology – still missing? The answer is computing power. We need resources that allow us to train neural networks quickly, enabling them to understand the world the way we, humans, do, but without massive energy consumption.
Today, we can develop computer vision solutions that excel in specific areas – detecting objects, identifying abnormalities in X-rays, detecting armed people in public spaces, and extracting subject lines and text from emails. This list can go on and on. However, even these solutions require enormous computational resources. To achieve human-level recognition and detection, we need thousands of GPU hours, which can cost tens or even hundreds of thousands of dollars. Not every company can afford such expensive neural network training experiments.
Overcoming Resource Barriers to Scale AI Solutions
Right now, only major corporations like Google, Microsoft, Salesforce, and AWS have the resources to train large-scale models. But that is the situation today. To be fair, I should mark OpenAI which emerged and grew thanks to its solutions and technologies, becoming a driving force in the GenAI area.
Are DeepML technologies and neural networks a dead-end? We don’t know. This is the technology we have today. Whether we will eventually hit fundamental limits, even as quantum computing advances and becomes accessible on portable devices, – remains uncertain.
