«To date, AGI is a good topic only for the Hollywood film industry».
Massimo Chiriatti, Chief Technical & Innovation Officer of Lenovo thus dismisses those who maintain that the advent of Artificial General Intelligence is now imminent, that is, an AI capable of understanding, learning and applying knowledge in different contexts as a human being would.
And then he adds: «Scientists clearly say that we cannot predict if and when we will make the numerous and enormous scientific advances necessary to create it».
But if an artificial intelligence with human capabilities is still far away, generative AI is already here, thanks to the success and popularity achieved after its debut on the global market with the launch of ChatGpt 3 (in November 2022), it is already being used to “increase” people and processes within companies.
But what is the difference between "traditional" AI, already widely used in companies, and generative AI?
«In theory there is no difference, in practice there is: we must make people scared by these innovations understand that in reality we have already been using these technologies for some time, for example when they use ADAS (Advanced Driver Assistance Systems) for driving assistance, equipped with sensors such as cameras, lidar and radar.
These systems act automatically in the event of danger, for example by braking if the driver appears to be asleep, and it is a well-established technology that physically intervenes through actuators.
If we move to generative AI, instead of receiving input from sensors it is fed with texts, code or images and generates new content, in the form of words and images.
In short, AI has gone from predicting obstacles along the road to predicting the next "token", the next piece of content, significantly expanding its fields of application.
In business, generative AI is already being used to save time, one of the main human motivations for using technology.
Applications range from the automation of information in the office to the reduction of repetitive physical tasks in the factory, from forecasting the quality of products to forecasting the maintenance of production machines.
This not only helps optimize processes, but also provides managers with crucial information for timely decisions, with the ultimate goal of saving time, energy and money."
With generative models, AI goes from being a tool to a work "companion". However, there is the risk that his ability to create (or at least, to simulate creativity by rehashing things created by man) calls into question the primacy of the human being and debases his abilities. How is this risk managed?
«When we abandoned horses for cars, everyone feared for the future of blacksmiths.
Yet, no one had foreseen the birth of new professions such as mechanics, tire specialists and auto electricians.
Of course, the car brings with it dangers and pollution.
However, we have built roads, developed safety technologies and introduced civil insurance to meet these challenges.
Of course, at the moment it is impossible to imagine what work our children will do, but I am sure that no one will ever want to do jobs where they can be replaced by a machine.
For us, technology only makes sense if it is used to "increase" human capabilities and, when this happens in companies, they become more competitive, grow and win over the competition."
Today, AIs are “black boxes” that swallow and digest data to process responses in inscrutable ways. Research works to guarantee the reliability of models and create trustable AI, but it takes time and in the meantime current systems often give false or imprecise answers requiring continuous checks, they can violate copyrights, they inherit our prejudices from the data we have produced. Is it possible to ensure safe adoption of AI in the company?
«I start from a mistake that we human beings have been making for some time: saying "you are a machine" to those who do an impeccable job.
It is a mistake because, in doing so, the idea was established that the machine could be perfect, infallible and now, with AI, this idea has become even more strengthened.
However, this is not the case: artificial intelligence can do a lot, but we must not take an output as an unappealable verdict and we still need to work to build verification systems that are truly capable of guaranteeing the reliability of the results.
So what we need is common sense: there are already many applications in companies, and artificial intelligence is already a precious tool for analyzing data and finding correlations that humans are not able to see, but it must be used with common sense.
To return to the comparison with cars, when we started using cars in the 1900s we knew they were dangerous, but we chose to manage the risk, and today they are infinitely safer."
Let's talk about environmental impact: ever larger models require ever greater computing power, with growing energy and water consumption and environmental impact. How do we solve this problem by simultaneously giving everyone access to these technologies?
«When we move an electron, or change a zero to a one, we generate heat.
This becomes a challenge when we process large amounts of data, well beyond human capabilities, and are increasingly immersed in tasks that require processing this "big data".
In this regard, companies like Lenovo are trying to improve the economic and environmental conditions for more extensive use of artificial intelligence.
A significant change, such as switching from air to liquid cooling systems, can lead to significant savings.
AI training can cost millions of dollars, but if you reduce the precision during inference, the process that allows you to use the model, you get significant savings in terms of consumption while maintaining a computing capacity suitable for the specific task.
So, our job is to first improve hardware cooling and optimize AI models.
And then we work on the ecosystem.
My IT architects design tailor-made solutions for the various workflows which include storage and edge computing: the latter allows compact systems to be brought where they are needed, ideal for areas where there are no data centers".
*Journalist, innovation expert and curator of the ANSA.it Artificial Intelligence Observatory
Reproduction reserved © Copyright ANSA