Artificial Intelligence (AI) is everywhere, and I don’t mean by that popular headline on Twitter or your favorite tech news portal. Let me show it here and now, while I am sitting and writing this article.
In fact I’ve just spoiled the first trend which is the fact that we often don’t talk about AI, it has become an invisible helping hand. I am even using it when I type this text right now. If I hear a notification sound, I will pick up my phone which uses AI to recognize my face. Another feature learns when I sleep and tries to automatically define my sleep routine to make it more organized. The watch on my wrist analyzes my pulse and suggests breathing exercises to calm me down and to relax me when it detects patterns of anxiety.
The text you are reading now is sent through the Internet using some kind of advanced software defined networks and routing which constantly learn and optimize themselves by using machine learning (ML). Of course, I see the advertisement on the page which uses my online profile to try to propose something that AI believes might be interesting to me. I yet have to open the package I received today, and its delivery logistics are probably more efficient due to AI optimization. I open my banking transaction list online and I automatically see assigned expenses categories, again based on AI.
If I decide to take a picture of the Moon outside my window, my phone will use advanced computational photography based on machine learning to create a nice night photo. It will almost magically stitch together multiple frames to create a beautiful image, despite the small lenses and the lack of a tripod, which is something that was completely impossible even just ten years ago. By the way, I can open my old photo now and AI will help me to do advanced upscaling and get rid of the noise, it can even remove unwanted objects automatically.
(night photo, handheld, bridge over Rhine, Cologne, GE)
For me, this is trend number one, AI is becoming more and more invisible and ubiquitous. There’s no “AI on/off” button. It’s helping with our everyday lives and business activities.
At the same time, we are growing dependent on the sometimes invisible but useful benefits of AI.
Let’s see what else is happening in the AI world right now and what the next steps are.
AI was entirely a scientific domain for a very long time and because of the limitations of hardware and software the majority of businesses could not benefit from it. There were “AI winters” after new hopes faded away and “AI winters” again and again.
However, since the 2010s, the benefits of AI have become a worldwide phenomenon, and thus the optimism was on the rise.
AI was supposed to save earth from asteroids, climate changes, all the businesses from economic depression, detect illnesses, and autonomous cars were supposed to fill our streets in 2020, at the latest. Images of robotic humanoid heads filled our web browsers, creating an overblown impression of what AI could already do and it generated a kind of AI bubble.
And then, we reached the peak of all the inflated expectations, which we addressed in the following article:
→ Is Deep Learning hitting the wall?
What is universally perceived as success both by machine learning experts and businesses, is that both sides finally learned what they can expect from AI, when to use which technique, how to achieve their goals, or to stop a project in its infancy to avoid great disappointments.
It is already helping to improve the success rate of AI projects. The current set of technologies and patterns are not miracles from some science fiction books or movies, they are much less fancy, but still very useful when implemented properly.
With AI initiatives, often comes the realization that the current state of the entire data management and governance is in need of major improvements. Data quality and simply the lack of representative data are recognized early and addressed more efficiently.
→ Read more It’s all about Quality: Migration and Data Validation Testing
There’s much less jumping over obstacles right into ML models and pretending that it will work somehow . . . it won’t, or at least not with sufficient accuracy, so it’s useless in the end.
This common understanding of what data teams and businesses can achieve together, today, and what is not yet possible, is a great and valuable lesson.
AI is no longer acceptable as a black magic box that makes decisions and generates output without the ability to answer WHY. This is especially important because our dependency on AI is growing fast.
For instance, why did the ML based model decide that we should be denied an increase of our limit on our credit cards? Or why was this road picked over that road in the online navigation?
If we depend on it, we want to feel safe and to know why; so do the data scientists. They don’t want to create some kind of digital AI monsters that get out of control.
XAI seems to be still a novelty but quickly is becoming a norm, as it is a standard requirement for advanced data projects, when ML is a key for decision making.
We already addressed this topic in a separate article “ Explainable AI (XAI) is what business needs in its path towards Responsible AI”.
Businesses feel the pain of the shortage of AI talent but also want to empower their people with a deeper domain knowledge in order to create and use their own ML models.
All the major AI technology vendors invest in simplified technologies, much more automated, and deployed in the cloud, so as to encourage citizen data scientists and professionals to join. It is helpful for simple cases, but it does not generate acceptable results for more complex cases, which unfortunately too often means … real life cases. Nevertheless, Auto ML keeps on improving and delivering better results.
→ Explore Citizen data scientist – low code movement equivalent for data space
This topic is often present in our Avenga Magazine, for a reason.
→ Natural Language Processing: Tasks and Application Areas
The main reason is, as the name implies, it is a natural way of communicating among people doing business and conducting their private lives. There’s a lot of business data available only as text. And, there’s an irresistible temptation by all businesses to automatically discover the meaning of all that data to improve efficiency and discover new patterns in business.
We are also observing an arms race in NLP models, as we remember our fascination with how natural the GPT-3 responded to our queries with human-like responses.
There are still many unsolved problems, with the context of the conversation as the most pointed out one. But, the technologies definitely keep on improving. It’s worth noting that GPT-3 came as a surprise, as it was thought before its release that a major breakthrough had been needed and progress had stalled. Now, we see new models being developed which are expected to surpass GPT-3 in a major way in the near future.
On the sour side of things, the expectations for voice assistants are still not being met and the progress is slower than expected by users. Even the leading Google Assistant needs more work, not to mention Siri from Apple which still hasn’t met its goals, to say it politely. “Siri’s jokes are not funny at all, but Siri’s bugs are really funny” said my nine year old son and I couldn’t agree more. There are billions of users expecting voice assistants to get better. They are key components of the touchless user interfaces of the future.
Data scientists used to live in their own bubble, somehow disconnected from the rest of the software process. Their work is still more experimental, as there’s a lot of trial and error because their processes are different. Different, definitely, but the AI/ML process automation is here to help speed up the testing of different models, as well as parameters tuning, in order to make it more parallel by using different infrastructures (more often cloud with its dedicated hardware and pay as you go models).
Computational notebooks have become a controversial topic in the face of MLOps and DataOps.
The AI popularity started with pattern recognition, for images, videos, texts, and our voices. This is where current AI technologies shine and are definitely great tools that are used almost daily in any business area. And, this set of technologies and practical applications will only grow, and AML, KYC, fraud detection, medical image analysis, etc. will all benefit from the improvements of analytical AI.
But now, AI has finally become very useful in its creative role. Which is why we dedicated an entire article to this subject.
We don’t need to look far to experience the benefits of generative AI, as many of the modern TV sets have automatic upscales and framerate enhancers which are practical examples of generative ML.
We can make old photos alive by just posting a static image of the person, which at the time of the writing of this article is very trendy online.
For businesses, it means the ability to simulate and forecast the future and prepare for different scenarios. AI can generate behaviors based on the past and parametrization, which helps to optimize strategies and processes for more resilient enterprises. Which, by the way, is a key trend for 2021.
On the darker side of AI, we cannot ignore the rise of deep fakes, which are becoming much harder to detect. Fake photos and Twitter posts are an old story, but the new story is fake videos produced in real time, replacing one face with another. It’s another example of how powerful technology can be used to benefit people or to harm them.
Despite the common belief and the Terminator/Space Odyssey 2001 movies that inspired fear, AI is more and more regulated by law. The importance of AI has been recognized by the European Commission and other major institutions and new legislation is coming.
The old question of who will be responsible for damages caused by autonomous cars will be regulated very precisely before any such cars may be widely seen on our streets.
Of course, there are the voices of disappointment with the current state of law, urging governments to step up their legislation efforts to protect their citizens from the wrongful use of AI against them.
→ Have a look at Human digital twins. What are they and Why are they? New hardware architectures
AI is supposed to hit the limits of computational power and energy efficiency. We saw AI moving away from traditional multi purpose CPUs to GPUs, then dedicated GPUs just for ML, and later TPUs, yet another major breakthrough is needed.
Quantum computing will allow AI algorithms to run much faster than with binary computers. We are already observing the first wave of major breakthroughs.
Neuromorphic computers, built more to resemble electronic brains, are also much more effective at training natural networks because their physical architecture is very close to the brain’s structure of neural networks.
→ More about Neuromorphic computers – moving towards supercomputers in our heads
AI, as I mentioned at the beginning, is in our smartphones, our watches, fitness trackers, cars, etc. Virtually any device with sufficient power can run some version of lite models to recognize images, heartbeat patterns, our faces, voices, and even steps.
Read about Trends in edge AI
As businesses learn to manage their expectations of Machine Learning projects, we will experience more successful digital initiatives with more predictable outcomes. No miracles are to be expected, however more valuable outcomes will be delivered.
The library of technologies, data, tools, and models is growing, so there’s a lot to pick from to accelerate new AI projects and to achieve better results in existing ones.
AI hasn’t lost its magic, even though we know much better now how to use it and what threats are related to increasing the proliferation of AI.
Many experts are predicting another major breakthrough in AI. There’s so much effort being spent on the development and popularization of AI that it can happen at any moment. But we shall not wait and see, because there’s so much that AI can offer for your business today.