My team at AI Newsletter, our weekly newsletter covering artificial intelligence, reached out to 17 of the smartest people in AI to find out what trends they see happening within the space over the next 12 months. Here’s what they had to say:
The true AI is the AI behind the AI that is thrown around as marketing buzzwords. When systems become sophisticated enough to be both self-aware and interested in protecting its creators and its environment. That is when AI will truly contribute towards mankind. Not as a technology but as a supporting actor for science.
Developer tools explosion in the area.
Agent-based models become mainstream.
The rise of AI in the sales industry is the trend to watch over the next year. It’s an important moment for salestech as we transition from cloud-based computing to AI. The prospect and customer information that salespeople use to benchmark everything they do has too often been subjective, incomplete, error-prone or just too complicated to capture.
Now AI-powered help is here and unlocking so much untapped sales potential, including sales coaching, lead generation, deal health analysis, enablement and insights.
This new era is made possible thanks to two critical innovations:
1) Advancements on our ability to automatically capture the full spectrum of business activity data rather than relying on manual data entry
2) AI’s power to separate signal from noise and surface insights that change the game when it comes to functions like pipeline inspection and sales coaching
The two combine to increase the amount of revenue reps generate year over year. We’re only scratching the surface, though. I wouldn’t be surprised if five years from now we will see AI-based systems partially or even completely running the sales process.
AI based copywriting is going to absolutely take over in the next 1-2 years. We’re already seeing a swath of GPT-powered software take over tasks such as headline writing, or copywriting. The genius of these systems is that they are also providing further training data.
The human user sees perhaps 5 generated headlines to choose from, the one they pick sends a reinforcement signal back to the AI to do more like that. Over time these systems will do copywriting tasks that are indistinguishable from human copywriters.
Perhaps the next iteration beyond that, as these AI models become faster, is a headline dynamically created for every single reader using clickstream analytics to determine what works best with any particular site visitor.
Humans will turn their attention to “why” AI makes the decisions it makes. When we think about the explainability of AI, it has often been talked about in the context of bias and other ethical challenges. But as AI comes of age and gets more precise, reliable and finds more applications in real-world scenarios, we’ll see people start to question the “why?” The reason? Trust: humans are reluctant to give power to automatic systems they do not fully understand.
For instance, in manufacturing settings, AI will need to not only be accurate, but also “explain” why a product was classified as “normal” or “defective,” so that human operators can develop confidence and trust in the system and “let it do its job”.
More implementation of AI to solve real humanity problems, real social problems, and in case of Blue Sky Analytics, problems around Climate Change, Environmental Degradation, and Sustainable Development. That’s the major trend we see.
Brands will turn to AI to help them better understand consumer conversations. The pandemic has changed how brands and consumers interact leaving many marketers in the dark. AI will help brands connect the dots between consumer conversations on social media, survey data, customer service transcripts, and many other channels.
The brands that quickly act on insights are the ones that will build strong connections with consumers. This will lead CMOs to look at more efficient processes and workflows to democratize data within their organizations.
Developing AI becomes mostly about training data creation and curation. Emergence of new class of data-centric tools and workflows.
Subscribe to AI Newsletter, our free weekly newsletter covering all things AI →
The biggest trend in AI during the next year will be AI safety. No, this doesn’t mean we should be concerned about superintelligent AI systems taking over the earth. Quite the opposite. Existing techniques in AI are able to perform really well on specific tasks 90% of the time, but may perform really poorly 10% of the time.
But, the 10% of the time the algorithm performs poorly can lead to really bad outcomes: accidents for self driving cars, language models that parrot common prejudices, or medical diagnostics that don’t work. Therefore, as more people work on applying AI to areas with low tolerance for failure, we’re going to need new techniques for building robust AI-based systems that limit downsides.
I think AI reached a certain maturity, it’s just getting started still, but a lot of people now know what AI can do as well as what it can’t do. There is a huge trend of training on very small data sets (on very few samples). In real life, it is very common to have less data, not more. Adversarial networks, GANs and generally speaking the ability to create smart synthetic data to be used for training is another interesting concept and a good deal of data scientists are working around this topic.
I think we are at an interesting juncture in AI with some exciting techniques starting to gain some traction in the practitioner community. There are two algorithms in particular that I think may be getting more attention, both of them having a strong connection to pure mathematics.
Firstly Topological Data Analysis, which exploits algebraic topology, allows the comparison of the “shape” of data in a non-trivial sense. It has found application in many areas as a classifier, but also in the detection of periodicity in time series.
Secondly neural net technology is usually posed as a discrete computational problem with neurons firing others according to signal weighting which is adjusted against training data. There are some recent developments recasting the problem as a continuous vector field, essential driving the number of neurons to infinity. This has promise as a way of growing the predictive power of AI and also allowing time dependent AI to become more of a possibility.
Outside of Algorithms, it is undoubtedly the case that ethics, bias and accountability are key themes for the next phase of AI. I suspect that we are not far from some attempt at regulating the bias in AI and the industry will start to see the emergence of lots of technological solutions to monitor the AI equity issue. It is a serious matter for the industry if we are to get wider adoption of AI outside of embedded and essentially hidden applications.
Artificial intelligence is the quintessential partner to build technology and software that make digital banking possible. With AI, human errors are at nil, making banking more intuitive, with enhanced customer experience. Additionally, banks will have opportunities to serve customers better while accessing technology to create new ways to interface with financial information.
Accelerated digital engagement has caused tectonic shifts in customer behaviour and soaring expectations. In banking, we could foresee trends like AI-enabled video-bots that impersonate an actual human or a branch officer and answer customers’ questions verbally, in a language of their choice. This technology will be made possible with advances in 5G, 3D human-like models, Natural language processing (NLP), AI, and finally, digital voice assistance services.
These AI-enabled video bots can completely transform the text-based chatbot experience while replacing branch banking for most day-to-day transactions. Imagine a human-like animated character, powered by AI, performing just like a traditional branch-based teller or executive to interact with customers.
Self supervised learning as cleanly tagged data is very expensive and it’s extremely hard to continue maintaining the process.
Artificial intelligence in sleep technology is going to take off in 2021. Specifically as it relates to advancements in wearables as more people, companies, and academic institutions start to dig deeper into this critical third leg of the health stool (alongside diet and exercise).
Dozens of new technologies are launching to help people better track and improve their sleep quality. For example, EnsoData’s FDA-cleared AI technology is already supporting sleep clinics’ shift to telehealth appointments, increasing access to sleep testing and diagnoses to improve patient outcomes for the millions of people who suffer from obstructive sleep apnea (OSA).
With advances in natural language processing, more companies are using artificial intelligence to quantify what previously felt qualitative – human experience. At Diversio, we use NLP to analyze free text data from employees on a set of 27 Inclusion Metrics. This allows companies to look at differences in experience between employee groups (for example, men vs women) to identify bias or harassment in their organization.
This is a significant step forward toward overcoming unconscious bias. When leaders are able to put numbers around experience, it enables them to set targets and accountability. It is no longer an “HR problem”, but rather a business priority.
AI is here to stay and grow together with us. We are collecting more data, and the AI improves over time. In the next year, we’ll see AI recommends us what to eat, what to buy and how to drive. In the next 5 years the AI will take a major part of our life and will help us to do things on our behalf.
Are you interested in getting access to exclusive interviews with AI experts, along with a recap of what’s happening in Artificial Intelligence every week? If so, subscribe to AI Newsletter, our free weekly newsletter covering all things AI.