Wildfires, the COVID-19 pandemic, and Brexit were only three of dozens of events that have made 2020 a year to be remembered. But apart from the social and economic tensions, research on Artificial Intelligence (AI) has not halted – in fact, this technology has changed the world in a way unseen before and 2021 will bring even more disruptive and amazing changes.
Recent years have seen a boom in the AI and Machine Learning (ML) market, with data science-based techniques changing day-to-day work in multiple fields, be that in retail, healthcare, or the automotive industry – you name it and AI has made its mark.
According to Markets and Markets estimations, the AI market is predicted to reach $4,127.2 million by 2021 with a Compound Annual Growth Rate of 33.8%. This impressive growth is fueled mainly by automating and augmenting existing workflows. Machine learning is already earning millions of dollars for Netflix and Amazon by delivering a first-class recommendation engine. Google delivers more accurate search results and filters out spammy websites by using a sophisticated application of Natural Language Processing called BERT (Bidirectional Encoder Representations from Transformers).
But these are just examples. The trend is something more – it is a direction seen in reality that empowers us to predict the future. Considering that, the Tooploox AI team identifies the trends and hot topics as the following:
- Natural Language Processing going mainstream
- Autonomous Vehicles reshaping the landscape
- Further AI impacts on society
- Synthetic media and deepfakes
- Hyper-automation and deep analysis on the rise
- AI big time in healthcare
- Edge AI – abandoning the safe space in cloud
AI Trends 2021 #1 – Natural Language Processing (NLP) still on the rise
Natural Language Processing is the set of techniques that enable machines to communicate and analyze information stored in an unstructured form, be that a press article, a report, an email, or an analysis like this one.
Currently, people are interacting with computers via artificial languages like Python or C++, or through sophisticated interfaces delivered by product design teams. With the rise of NLP, it is increasingly popular to interact with machines in a way humans consider convenient.
The year 2020 came with the next iteration of the Generative Pretrained Transformer (GPT) natural language processing model as delivered by OpenAI. The Elon Musk-backed organization delivers increasingly sophisticated models that can produce texts of a quality previously unseen by machines.
The text produced by the model is in fact indistinguishable from ones delivered by human writers, making the technology potentially dangerous. In the wrong hands, the model could be used to spread fake news or extremist propaganda.
The model has been described in an Arxiv paper and further has been licensed exclusively to Microsoft – a highly controversial decision when it comes to AI ethics. For public use, the model was delivered as an API.
AI Trends 2021 #2 Autonomous vehicles – Robots will not catch COVID
Autonomous vehicles seem overhyped, with no driverless cars actually driving through the streets, yet with multiple headlines fueling public interest. The COVID-19 pandemic has significantly reduced that interest due to more urgent matters needing to be solved. https://trends.google.com/trends/embed/explore/TIMESERIES?req=%7B%22comparisonItem%22%3A%5B%7B%22keyword%22%3A%22autonomous%20cars%22%2C%22geo%22%3A%22US%22%2C%22time%22%3A%22today%2012-m%22%7D%2C%7B%22keyword%22%3A%22driverless%20cars%22%2C%22geo%22%3A%22US%22%2C%22time%22%3A%22today%2012-m%22%7D%5D%2C%22category%22%3A0%2C%22property%22%3A%22%22%7D&tz=-120&eq=geo%3DUS%26q%3Dautonomous%2520cars%2Cdriverless%2520cars%26date%3Dtoday%2012-m%2Ctoday%2012-m
But this doesn’t mean the research has stopped. Not even close. Ford initially predicted that there will be fully autonomous vehicles on British roads by 2021, yet the company later postponed their plans to 2022. The pandemic has hit severely, indeed.
But research continues and even if we will not see autonomous vehicles on the road in 2021, we will see them elsewhere – in the fields or on the rails.
Road traffic comes with multiple challenges unseen in a more controlled environment – the driver needs to control other users of the road, maneuvers, or any unexpected situations, like children running onto the road.
When it comes to railroads, the number of factors to control is significantly reduced. Railways are usually more separated from populated areas, or at least fenced. Trains find it hard to take turns – and so on.
According to International Energy Agency’s predictions, rail transport has a significant chance to cut emissions caused by other types of transportation, especially road traffic and short-distance flights. Thus, there can be a rail revolution looming in 2021, yet this time powered by autonomous locomotives, especially considering the fact that there are already running prototypes.
AI Trends 2021 #3 Societal impact of AI
Artificial intelligence and machine learning-based technologies are going mainstream in culture and society, bringing their own set of advantages and challenges. But it is not a surprising trend at all.
In fact, it is common for every emerging technology to inspire culture and impact society. The process has been seen when steam trains revolutionized transportation, and so it is with AI and ML trends bringing their own impact.
With the growing impact of AI-based solutions on people’s daily lives, the need for ethical boundaries are rising. Also, a Machine Learning solution is no better than the dataset it was built with. And that brings more problems than one could ever expect. More about these challenges has been covered in our recent blog post about AI ethics and datasets.
AI Legal Challenges
Last but not least, the impact of AI-based solutions requires updating our legal framework. There are also more visible challenges concerning the responsibility of creators or the interpretability of the AI-based solutions. Already there are works from various institutions on the regulation of Machine Learning solutions, including:
- The European Union has released whitepapers about building ethical AI solutions
- The NeurIPS conference encourages researchers to predict possible outcomes of technologies
- The Vatican has issued a pledge about ethical AI
The year 2021 will hopefully be the year of transferring the advice and guidelines into full-fledged legislature. In fact, despite the tremendous growth and unprecedented impact on people’s daily lives, AI solutions were some kind of no-man’s-land when it came to its legal situation – more details about the matter can be found in our blog post about AI through a lawyer’s lens.
AI Trends 2021 #4 Synthetic media and deepfakes
Artificial Intelligence is, as a tool, transforming art as unseen before. There are tools aimed at automating creative workflows by reducing the amount of repetitive and dull work. But there are also more sophisticated approaches.
Apart from technological conferences focused on improving and analyzing these tools. There are also art-centric initiatives like aiarthists.org.
Apart from the art made by human artists, there is also fully synthetic media, where the outcome is fully automated and used to enrich traditional media productions. Or to create (and re-create) reality in a way the designer finds most suitable.
This comes with multiple outcomes – starting from delivering artificial music, art, and even artists. With AI-based solutions, users can produce fake voices or deliver images of people saying whatever the designer wishes. This is commonly associated with deep fakes – a sinister usage of neural networks that can be used to produce videos of politicians and celebrities saying things they would never say.
On the other hand, though, deep fakes enabled Star Wars fans to see Grand Moff Tarkin on the big screen once again, as in Star Wars: Rogue One. Delivering the effect without this technology would be challenging – as Peter Cushing, who was seen in the original A New Hope as Grand Moff Tarkin died in 1994 aged 81.
But quite frankly, he didn’t appear all that dead in the movie:
A strong hint on the significance of synthetic media in the AI Trends of 2021 is an increasing number of interesting startups harnessing the power of this group of solutions. To name just a few more interesting ones:
- Comixify – the company enables a user to rapidly change video into a comic book
- Sensity – this startup aims to reverse-engineer the near-perfect deep fakes using deep learning technology and automate the flagging of fake news
- Rosebud – the company delivers AI-generated models which can be either depicted on an image or in a short animation, effectively delivering an infinite number of models with different faces.
AI Trends 2021 #5 Hyper Automation and deep analytics
The main goal and the key purpose of machine learning technology is to bring automation into the workflows that used to be impossible to automate. With image recognition or Natural Language Processing, building AI-powered, humanless business processes became possible.
This trend is only going to accelerate. With the wider availability of AI technology and the growing possibilities delivered by it, there are more and more use cases. Just from Tooploox’s experience – it is possible to deliver an automated meal recognition system that tunes oven settings to deliver a dish cooked to perfection.
Personalization and recommendations are nothing big in 2020 and 2021 – they are a must. Now there are even attempts to deliver cashier-less stores that harness the power of image recognition to deliver a smooth shopping experience. After Amazon delivered the first concept store, the French startup Storelift came up with the next attempt.
When it comes to the hyper-automation trend, it is hard to say anything for sure – it is about pushing the boundaries of already existing technologies to an extent unseen before and it is probably the strongest AI 2021 trend among those listed.
AI Trends 2021 #6 AI big time in healthcare
According to PwC data, the market for AI in healthcare applications is estimated to reach up to 6,662 million in 2021. It is not surprising, after all. Medicine is not a hard yes-no science, but a statistics and experience-based field, in which looking for the perfect cure or spotting signs of a disease come from dozens if not thousands of trials and observations. Also, AI can be introduced to healthcare from various platforms including cloud, on-prem and mobile healthcare apps.
As an offshoot of this paradigm, healthcare produces tremendous amounts of data – which can be a blessing and a curse. The sheer amount of produced information, gathering it into one system, managing and, finally, storing it effectively is a challenge in and of itself, one much better explored in our blogpost about Electronic Health Record and AI.
Yet there are two ways in which AI will support the healthcare industry in 2021.
Automating Daily Non-Medical Tasks
Apart from physicians, nurses, and the medical personnel, healthcare is about the piles of paperwork left to process by administrative departments. AI can provide a deal of support when it comes to processing the tedious and manual tasks of healthcare specialists, effectively freeing their time and reducing the overall cost of providing care.
Supporting The Medical Tasks
A large portion of medical data, including MRI scans, x-rays, and CT-scans take the form of an image. AI-based solutions are currently of high proficiency in image recognition and processing. Considering that, it is natural to support the diagnostics with image-recognition based solutions. Virtum, as delivered by MicroscopeIT, a Tooploox-partnered company, is a perfect example of using AI in histopathology, effectively aiding analysts in spotting cancer cells in tissue samples.
With the advancements in shaping the legal framework and a growing social consciousness about AI-powered solutions, the year 2021 will witness the further development of AI healthcare solutions.
AI Trends 2021 #7 Edge AI
Bandwidth and data transfers have been one of the bottlenecks in new technologies for a long time. Delivering better internet and faster transfers is just one way of tackling the problem.
Yet there is another way, delivering even better results – AI on the edge, where a neural network is doing all the dirty work right there on the device, instead of transferring all the data from the device to the cloud.
Thus, when placing a neural network on an edge device, one can cut the transfer to only the effect of the work – sometimes it can be just a trigger for other devices. Sometimes the transfer can even be made entirely unnecessary, when the job is done on-site.
With the development of all the technologies mentioned above, the need to process data on-site instead of transferring it around the globe will be growing – especially considering the fact that there can be multiple unexpected events. Due to the COVID-19 pandemic and the lockdown, YouTube and Netflix decided to reduce the quality of streamed movies to keep the internet infrastructure operational.
When faced by Netflix and Youtube, this challenge seems manageable. But if the quality of healthcare had to be reduced, it would be a much more challenging process. And the pandemic has reminded us that it is impossible to predict everything.