https://www.youtube.com/watch?v=pKi0mK84MNk
What were the biggest AI topics and trends of 2020, how about the latest AI predictions for 2021? Stick around for the first AI News year in review to learn more. Hey everyone, I’m Alex Castrounis and this is AI News the number one show for the latest topics and trends in artificial intelligence.
Subscribe to this channel to get the latest episodes of AI News and let’s dive in. For this year in review, we’re going to recap a lot of the developments, trends of 2020 in artificial intelligence as well as talk about what to expect in AI in 2021.
Be sure to let us know in the comments of anything from 2020 that we missed or anything from 2021 that we should expect that we didn’t talk about today. To kick things off, let’s talk about the elephant in the room, which is obviously that 2020 uh was the year of the Covid pandemic, which had major major impact globally on people’s lives and everything else.
Specifically when we’re thinking about artificial intelligence it also had an impact there and one of the things is that because of the pandemic and the way it sort of reorganized how people work, and remote work, and, you know, the ways companies have had to adapt, it may have accelerated AI adoption to some extent.
Also, it might have, you know, impacted company data sets significantly. So, if you think about it, you know, companies often use their data to analyze, you know, what happened in different quarters throughout the year or just throughout the year in general and maybe try and make data-driven decisions based on that or see kind of how how their customers are buying or any changes and trends, whatever the case may be.
Obviously, due to this pandemic, it’s had a major impact on everything across the board from people’s purchasing, where they purchase, how much they purchase, if they purchase at all, and so many other things and so in some ways the data that was collected, generated and collected, throughout the year 2020 is sort of very skewed obviously.
You know, think of it like how seasonality and certain, you know, holidays or whatever might affect data and this is sort of a similar thing but on a, you know, bigger scale obviously. So time will tell, you know, how Covid really impacted um, you know, the field of artificial intelligence, but um certainly it has to some extent.
So starting off with advancements and trends in AI specifically, we’re going to go through 2020 stuff and then again move on to 2021 and talk about what to expect for the future. So in terms of advancements and trends one of the biggest advancements uh in artificial intelligence in 2020 was really in the areas of natural language techniques like natural language processing, or NLP, and natural language understanding, or NLU.
One of the big developments there was that so-called transformer models really took center stage and sort of propelled the natural language um capabilities further than ever before. And one of the most notable and recognizable examples of that is this GPT-3 model that OpenAI created and sort of released to the world and sort of wowed everybody.
Also, can be considered kind of scary by many as well, and again let us know in the comments what you think about some of these capabilities that we’re seeing nowadays with models like GPT-3.
But we’ve also seen new benchmarks in natural language processing and understanding, you know, previously benchmarks like GLUE and SuperGLUE have been sort of the the industry standard um and then you know now we’re starting to see as these models get super good at dealing with these benchmarks at like human-level intelligence you know people are trying to create new benchmarks that really push this area of AI even further.
Another area in natural language that’s really come a long way is voice recognition and assistance. So, you know, voice recognition specifically in, you know, speech to text applications and so on.
So clearly we’ve seen a lot of advancements there, you know, voice recognition technologies are getting better all the time so are the assistants and they’re sort of better able to answer questions, look things up, you know, and so on.
Another area which is sort of similar to GPT-3 and some of these transformer models is the use of techniques like transfer learning and fine-tuning. So previously these techniques were pretty widely used in computer vision.
For some of you that are familiar with this massive image database that you can use called ImageNet, you know, a lot of people in the computer vision area had been pre-training models on this massive sort of set of images in ImageNet and then, you know, doing what’s called fine-tuning to apply this pre-training or learning to their own specific domain or problems that they have and we’re seeing that now more and more being used in natural language processing as well including with things like those transformers and GPT-3 models and so on.
Another thing that really was big in 2020 was this idea of augmenting human intelligence over you know full-scale job replacement or job automation. What that means is that, you know, augmenting human intelligence is all about, you know, using AI to automate some of these, you know, repetitive, tedious, rote, menial sort of whatever tasks that people do day to day at their jobs.
Think of it like administrative stuff and that sort of thing um and so by automating some of that stuff with AI it allows workers to actually focus more on you, know, value ad type work be more and also be more productive, be more creative, really do what they do best and so uh in a lot of ways it’s kind of a win-win because, you know, it allows again people to really do all those things but also enjoy their job more hopefully because they’re not doing all these sort of menial and repetitive tedious tasks, but it also because it results more in employee sort of happiness and enjoyment potentially, it helps companies with, you know, employee retention, happier employees overall, and so on.
Now that being said, you know, anything like social or sorry um augmenting human intelligence or job automation, in general, can always have it’s, first of all, it’s very complicated the whole there’s a lot of nuance to it and you know the question of automation in general and you know whether it’s good bad or otherwise is very complicated and obviously these things can also have potential socioeconomic impacts as well including things like inequalities and so on so a lot of the focus then needs to transfer in many ways to you know reskilling, upskilling, training, and so on for people to be able to work in this sort of increasingly uh data powered and you know advanced technology type world that we live in where AI is one of these technologies.
So we’re hearing more and more about that in 2020 and I expect we’ll hear a lot more about that in 2021, definitely something worth keeping an eye on, of course. Let us know in the comments your thoughts on any of those things and if you’ve seen any of uh that in in the real world yourself.
Another area that sort of gained a lot of steam in 2020 was this area of AI called generative AI. So you know there’s sort of some silly novel applications of that you may have seen, what they call style transfer where you know maybe take a picture of someone and then transfer it to look like a Van Gogh painting or a Monet or something, so sort of transferring one style to an to something like a picture.
But the the bigger news headline in generative AI in 2020 has really been more around what’s called deep fakes. Specifically audio and video deep fakes, so sort of AI-generated audio or AI-generated images or video where you know nothing you hear or see is actually real, it’s generated by AI, and obviously that could have huge sort of negative adverse impacts on many things uh in society and so you know people are it’s certainly the advancement of that area has has definitely raised concern amongst many and uh increased you know we’re starting to see sort of increased regulations and attention being paid to that so uh that 2020 we saw more of that and again we’ll see what happens uh in 2021 and beyond um and of course, let us know your thoughts on on things like deep fakes, which clearly have gotten a lot of people talking uh recently.
Another area that’s been big in 2020 was reinforcement learning. So it’s not a new area of AI, but in terms of the applications going beyond just playing video games for example you know reinforcement learning as an area of ai really gained a lot of steam and notoriety at like winning games like chess and and go and you know more in the game area, but now we’re seeing applications in drug discovery in you know automatically finding optimal um what they call hyper parameters to find the optimal neural network architectures and things like that um and so on and so forth and a lot you know control systems different things so reinforcement learning has really gained a lot of steam as well in 2020 and so that’ll be an interesting area to keep an eye on as well so has automation and we talked about automation before, but really more automation as well in the area of what they call RPA or robotic process automation so basically taking like think of companies they have a lot of different processes that they you know operational processes or you know whatever different sort of routines or processes that they go through all the time um and so just trying to sort of automate different different processes um you know one at a time to kind of take care of those things using these different techniques like ai so my process automation has been getting a lot bigger as well.
And then finally um hardware types of advancements. Hardware, whether it’s you know computer chips that process you know are able to do the training of AI models or do the infra actual inference so when you pass data to them you know uh these things that hold the models and then process the data and create a an output or a prediction, for example, but as well as like computing hardware too you know.
Different um kind of you know now that everything’s really heavily in the cloud you know different kinds of cloud-based computing hardware or even what they call edge-based hardware like edge being things like mobile devices you know now more and more mobile devices like smartphones are running AI models right on the device.
And so then you’re starting to see like you know edge AI as well as embedded AI things like that um and so some of the hardware has really come a long way in terms of being small and cost-effective enough to you know do these edge cases and embedded AI cases as well as really take cloud computing to the next level uh and making it a bit more accessible, a bit less costly, and a bit faster to train you know a bunch of different AI models.
Moving on to AI applications and use cases you know we’ve seen a lot more in 2020 really across the board meaning in different industries and different business functions you know it’s really we’re really moving away from AI being this like largely academic thing to AI being very much deployed in production in the real-world.
In real applications that we interact with all the time whether those industries are you know especially and very interestingly seeing AI being used in real-world applications in healthcare, biotechnology, and pharma, to sort of benefit people right whether it’s just discover more effective drugs quicker or detect diseases earlier so that people can get earlier treatments and better health outcomes.
Or whatever the case may be we’ve really seen a lot of that which is which is a good thing in terms of thinking of AI as being beneficial and being used for good purposes. We’ve also seen AI being used benefit in beneficial ways to protect animals or environmental protection and things like that, so that’s a really cool area of AI as well.
And, in general, you know there’s a sort of more people talking about AI to benefit people um which I talk about a lot in my book, which is why my book’s called “AI For People and Business”, but also you know this idea of AI for good right or AI for humankind um obviously AI can be quite scary for a lot of reasons uh as well.
And so you know the idea of AI being used for good as opposed to bad is also related to this concept of AI safety which we we’ve seen in 2020 people talking more and more about. So the idea of you know again AI not being used for harm but actually for good.
So that’s something that that’s been happening that’s been uh pretty interesting as well. Now on the different AI tools and data front. We’ve seen a lot more tools you know develop, coming out, or being developed and advancing more.
So one of those areas, aside from the heavy hitters like you know TensorFlow or PyTorch or Scikit-learn, like these common tools that a lot of AI and machine learning engineers might develop solutions with we’ve also seen sort of an explosion in cloud platforms and APIs offered by major cloud platforms like AWS and GCP and Azure and different services.
Even deep learning as a service type platforms, so definitely a lot going on there in 2020 that’s been a big topic as well as data in general just seeing a lot of you know focus on the data that is used to create and train these machine learning or AI models, whether that’s techniques like data augmentation, which means maybe you have a small data subset and you need a lot more data so you use the data that you do have and you kind of tweak it and rework it a little bit so that you have more data.
So you augment your data with variations of the data that you have, or synthetic data where you just create data synthetically using algorithms or programs of some sort. There’s also been a focus on data democratization, so you know making data more accessible to people within the company or in general or to the public.
Sort of less silos more availability more accessibility and even things like so-called open data. And now there’s a lot of publicly available data sets as well including even data search tools like Google created a data search tool that you can just go to do a search and you can kind of see all sorts of data sets for any given sort of type of data that you’re looking for all from.
This one tool which is really interesting in another area in 2020 there’s been a lot of focus on sort of ethics safety and impact. I kind of talked a little bit about that before, but you know one of these ideas of again ethics like right versus wrong, good versus bad, using AI in those ways, but also this idea of kind of AI responsibility or responsible AI.
So you know not only using AI responsibly, but being responsible for the way that AI is used um i i mentioned the AI safety thing already, but you know there’s just been more focus on the ethics it’s making more news and headlines especially in the areas of things like algorithmic bias, AI fairness.
So you know AI that benefits people should benefit all people not just specific subsets of people. AI safety and then you know AI is one of these things that can have just like any technology you know big impacts potentially on society, economics, the environment, and so on.
So kind of keeping an eye on those things and focusing more on that. We’ve even seen frameworks that have come out to kind of address those specific areas that could be impacted by AI and how to kind of recognize that how to plan for it, how to mitigate any negative impacts, and so on.
And as a result too there’s many more organizations and partnerships, again different frameworks and models even you know guidelines and now even companies that are consulting on AI ethics and so on.
So that’s really been kind of a bigger thing in 2020 as well. And another area kind of related to that is really this idea of governance regulations and compliance. So governance you know, whether it’s corporate governance, like how companies you know manage the data and the applications and things like AI and so on, that they have and the processes behind it and how they you know protect privacy and implement security and so on, but also even that the government level across the world different governments how do they create laws and regulations um to protect people’s privacy let’s say security and so on.
And then the idea of compliance not only with those kinds of regulations but compliance as well with even internal company you know guidelines or best practices or other standards that aren’t necessarily laws or regulations per se but you know companies try and comply with them because they’re the right thing to do.
And as such just like AI responsibility or responsible AI there’s also this idea that we heard more about in 2020 of accountability and accountable AI. You know who’s accountable for when things go wrong or when you know compliance is not happening or whatever and so um you know we’ll see what happens.
There’s a lot going on there you know obviously GDPR is a big example of some of this in europe the us has some similar things mainly in the state of California. Canada has some regulations like that in other countries as well so um we saw more of that in 2020 and of course we’ll see more uh moving forward too.
And then finally in the 2020s or year in review there was another area that you know kind of gained some steam uh and also is worth looking at which is AI education and talent first of all there’s a talent shortage there has been that’s not new in 2020, uh but you know there’s definitely a talent shortage.
But what 2020 did do is you know a lot there’s a lot more focus now on training and on learning and you know different things like that so um we’ve seen a lot more increase in sort of educational offerings or training opportunities or whatever in artificial intelligence, machine learning, data science, and similar fields.
Also in terms of talent you know there’s definitely been more of a focus, and for a good reason, on diversity and inclusion and you know that’s been an issue in technology for quite some time and is especially an issue in some of these areas like AI as well and so it’.
s a good thing in my opinion people are talking about that and focusing on that more You know, of course, let us know in the comments of any experiences you’ve seen out there or thoughts on that but so we have seen more of a focus on that in 2020 and more news about that.
All right, now on to the 2021 predictions and what to expect moving forward especially next year and beyond so some of the things that people are talking about again in advancements and trends as the first category will kind of cover, I mentioned RPA or robotic process automation that’s certainly not going to slow down so you should expect to see more of that in 2021 and beyond and more deployments of that you also are starting to hear more this term hyper-automation.
So you’ll probably hear more of that as well, but it’s basically automation sort of automation using AI versus traditional automation the concept of automation isn’t new and isn’t necessarily AI-specific at all other people have been automating things for a very long time and pretty much throughout the entire industrial age really since the industrial revolution.
Especially more and more with software and algorithms writing computer code, most of the software we’ve used well before AI was really a thing or used widely in the real-world. A lot of things were being automated that we used or or that you’ve come across in companies uh and so on but you know now this idea of AI based or AI-powered automation sometimes called hyper-automation so seeing more and more of that.
Another area that’s interesting is simulation and this concept of digital twins so you know I talked about synthetic data before so one thing we’ll see more and more of is companies sort of creating simulations to auto-generate synthetic data but then also creating sort of replicas of their company or certain processes that their company goes through or even like a whole supply chain.
Let’s say you create almost like a computer simulation of all of that right what they call a digital twin and then you could run that simulation with different parameters and things like that so that you can kind of automatically generate data, but you can also see how different models and different you know sort of pulling different levers or making different decisions or whatever affects the digital twin before you actually implement that in the real world, so definitely something to keep an eye out for in 2021 and beyond.
As I mentioned natural language processing and understanding is a hot area of AI and will continue to be so expect advancements in that specifically automated speech recognition or ASR conversational intelligence, virtual assistants, chatbots, and you know.
Lately, people have been talking about a new version of GPT-3 called GPT-4, which may have one trillion or more parameters. For those of you that know what that means you’ll know that GPT-3 had billions and billions and billions of parameters and was a huge model and so you know largely the reason it made so much progress is sort of that size that you know the idea is that the more parameters you have the closer and closer and closer you’re getting to maybe how the human brain works and how many neurons and synapses the human brain has so keep an eye out for that see whether there is a GPT-4 and whether you know it sort of sets the new record for one trillion or more parameters.
Reinforcement learning, expect to see more there, more development, more real-world applications, and again branching out more and more from just you know playing games and beating games and then also keep an eye out for new techniques so you know self-supervised learning.
For example, I did a video earlier on my channel, you know a little while back on self-supervised learning. That’s a super interesting area of AI and it’s a whole different way of doing kind of learning as opposed to some of those other techniques like supervised or unsupervised learning, so keep an eye for that.
The other thing is you know a lot of companies don’t necessarily have that much data and you know we talked about the synthetic stuff or simulation but also there are these areas of AI that focus on what’s called zero-shot or few-shot learning so the idea that you can have models that could be trained on very few or no data samples, so basically having very little data, or no data.
Self-supervised learning is kind of like and when I say no data I don’t mean no data whatsoever but maybe no labeled data for example in the case of supervised learning. But anyway that should be an area to keep an eye on as well because there’s a lot of active research going on there and an interest in being able to create these models that don’t need much data at all to you know be usable and effective.
Another thing is you know if you’re familiar with this idea of descriptive analytics and predictive analytics so like looking in the past, looking at data, understanding what the data tells you from a historic perspective, then predictive analytics is this idea of taking that step further and making predictions using models like AI and machine learning models then the next step from there is prescriptive analytics so not only making predictions but also you know optimizing certain things based on the data you have or the predictions that these models can make as well as even recommending actions or recommending you know decisions or automating actions and decisions based on these sort of prescriptive AI models.
That’s going to be an interesting area to keep an eye out for and something you should expect to see more of, and some people refer to this as well as decision science. As opposed to just data science, if you hear the term, decision science think about you know that kind of thing where you know you’re finding ways to automate decision making or taking actions or whatever so on the tools and data side of things you know um over the years especially in software develop traditional software development and things like building mobile apps or SAS platforms or whatever the case may be there’s been a much bigger focus on what’s called ops, specifically on DevOps, so the idea of like you know creating these deployment sort of automated deployments of software and updates to um the software so when you’re releasing new versions of it or new features but also like building out infrastructures cloud-based infrastructures kind of automatically using tools like terraform and so on.
And that spawned this whole sort of new area of ops called data ops sometimes people refer to that as well as almost synonymously with data engineering so like creating data pipelines you know data back ends like data lakes, data warehouses, and that kind of thing as well as on the analytics side like ML ops or machine learning ops and AI ops so you know how you deploy these models, how do you version track them, how do you you know swap them out as you increase them, how do you monitor what’s called model drift so maybe over time the model isn’t performing as well or being as accurate because data changes and think you know conditions change.
So how do you monitor all that? How do you alert on that? How do you you know version control? All kinds of things you’re doing and so on so that’s definitely an area to keep an eye out for in 2021.
Gaining a lot of steam as well as a similar area called AutoML, which is you know sort of automated machine learning. The idea is, you know, right now a lot of how data scientists and machine learning engineers build AI and machine learning models, is by tweaking all these parameters trying a lot of different models, it’s very experimental, it’s very trial and error, it’s time-consuming you have to try a lot of different kinds of models and even within the same type of model tweaking that model, what’s called hyperparameters to find kind of the optimal model.
So the idea of AutoML is not only to help find those best performing models as quickly as possible, but even to automate some of the deployment of the models whether it’s deploying to a restful API endpoint or something that you access via the web or HTTPS or whatever the case may be.
We’re starting to see more and more of that being applied to sort upstream of that, which is more the data preparation side. So how do you you know, a lot of people will say like 80 percent of the time spent building AI and machine learning models isn’t actually the model training our model development part, but rather the data preparation part, or what they call data wrangling or data munching, so you know sort of automating some of that data preparation to clean the data, prepare the data, make sure all the fields and variables and and whatever are normalized the right way if they need that or what they call standardized and so on.
And then lastly in the data side of things, this idea of data democratization that I mentioned before in data availability that’ll be an area of 2021 to keep an eye on as well. We’re seeing more and more of that all the time.
We’ll see what the future brings as well in the area of ethics, safety, and impact that I mentioned before. Again, I think responsible and accountable AI is becoming much more talked about than previously and should continue to be moving forward as well as AI fairness and safety that we mentioned as well.
So keeping an eye out on like you know what are the new are we going to see more guidelines and standards and best practices on that stuff? Are we going to see more regulations, especially around privacy and security, which we’re already starting to see with things like GDPR and the California Consumer Privacy Act or CCPA? And then there’s even an area of AI called federated learning that is being much more talked about and should see advancements in in the future.
I talked about that in some previous videos, but basically this idea of sort of distributing the learning onto devices, like kind of the edge thing, so that there’s not one central database that could be potentially you know breached or something like that and kind of making almost like private models for individuals like you have your own cell phone.
And maybe your cell phone is training models that are good for you and work for you but aren’t accessible to anybody else. Your data is not being exposed anywhere else and so on. So that’s sort of some of those ideas of federated learning and 2021 should bring about more of that.
And as I mentioned around this idea of governance regulations and compliance, not only keeping an eye on some of the things I just mentioned but also there’s these concept of AI explainability, interpretability, and transparency as well.
And so, you know, there’s an explainability is the idea of being able to explain how an AI model works and how it makes decisions. One of the problems is you have these things so-called black box models like deep learning often uses these like, you know, very intense neural networks or you have random forest which is a type of decision trees and you know someone asks you well how does this thing work? How does it make decisions? It could be very hard to explain in plain English, if not impossible, because of how complex the models are and how they work under the surface and so this idea of creating models that are explainable or techniques to take these more black-box algorithms like neural networks, deep learning and making them explainable as well as interpretable.
So really being able to understand you know what specific factors are they prioritizing over others. Which parameters and which variables are having the biggest influence on the outcome of the model.
So let’s say you’re trying to predict something like a stock market price or whether you know an image has a cat in it, for example, you know. How do you interpret which things are having a bigger influence on that ability to accurately predict a stock price or accurately detect a cat in an image versus other things that might be in the model but not have that big of an effect? And then transparency, which is just really the idea of you know understanding like having access to see how these models are working, or what they’re being used for, or who’s making decisions around them, or you know basically you know making things more available and more open to whoever.
So other areas to keep an eye out for sure. And then finally you know the education talent thing category that we talked about for 2020. Again there’s a talent gap so I think you know this new focus on this idea of data literacy you’re hearing more and more about.
So helping people better understand, speak the language of data, and understand how to use data to do things at within their organization. Or to help their organization or to help people like we said sort of AI for good, that sort of thing and just seeing more and more offerings out there.
Educational ones, I have my own and so I do a lot of training, I do a lot of speaking, I wrote my book. I’ve been on sort of this mission to help people improve their data literacy more and more for a long time as well as AI literacy.
I have my own organization called whyofai, which you can find at whyofai.com. And the idea there is to provide sort of AI training and education to specifically to more non-technical traditionally, non-tech, or typically non-technical folks like executives managers and entrepreneurs, but also for practitioners that are more hands-on and technical but that want to look through the lens of business and strategy and things like that.
So that’s another thing but of course, there’s so much out there nowadays so you know we’re going to see more and more of that. So in general there’s been a lot going on in 2020 across the board not AI-related at all and then of course a lot in AI as well just like we talked about.
And 2021 is certainly going to be exciting in terms of seeing all these developments in AI across all the different categories that we talked about. Let us know in the comments like I said, what are your thoughts? Did we miss, did I miss anything for 2020? Are there other things that we should be looking at for 2021 and beyond that wasn’t discussed here today? One thing I think is worth mentioning is that AI has always had a lot of hype around it and so determining what is hype versus what is reality is not necessarily easy for a lot of folks, but also people tend to be overly optimistic often, I found, and so the hype, even the predictions and the pace of development and advancement tends to be a little overly optimistic compared to what we really see.
Autonomous vehicles are a great example of that, so we’ll see. Keep in mind who knows how quick things will move and whether maybe it is overly optimistic or maybe not. Let us know what you think in the comments.
All right well that’s it for the first-ever AI News year in review. Hopefully you learned a lot about what happened in 2020 and also what to expect in 202. Be sure to subscribe to this channel if you haven’t already and check out the description below for more information and resources to help you along your data and AI learning journey.
Thanks again, and I’ll see you in the next episode of AI News.