Main 2020 Developments and Key 2021 Trends in AI, Data Science, Machine Learning Technology
Our panel of leading experts reviews 2020 main developments and examines the key trends in AI, Data Science, Machine Learning, and Deep Learning Technology.
Earlier, we covered AI, Analytics, Machine Learning, Data Science, Deep Learning Research Main Developments in 2020 and Key Trends for 2021.
COVID has changed how people live and work, and also highlighted the dependence of predictive models on assumptions and vulnerability to sudden changes.
This blog will focus more on Technology. We asked our panel of experts:
What were the main developments in 2020 in AI, Data Science, Machine Learning, Deep Learning Technology and what key trends do you expect in 2021?
Some of the common themes were AutoML, Automation / RPA, Bias in AI, COVID Impact, Deep Learning limits, Ethical AI, GPT-3, Medicine and Healthcare, and MLOps.
One huge AI/ML development happened just after the cutoff date for submissions. AlphaFold from DeepMind has apparently solved the 50-year problem of protein folding, with huge potential for medicine and biology.
Here are the expert opinions, with contributions from Marcus Borba, Kirk Borne, Tom Davenport, Carla Gentry, Jake Flomenberg, IPFConline (Pierre Pinna), Nikita Johnson, Doug Laney, Ronald van Loon, Bill Schmarzo, Kate Strachnyi, and Mark van Rijmenam.
Marcus Borba, Global Thought Leader & Influencer: #AI #MachineLearning #DataScience #BI #BigData #Analytics #IoT #DigitalTransformation.
In 2020, the growing concern to understand how AI works, and also assessing the issue of bias, make the need for AI to be increasingly explainable and responsible. The use of AutoML has brought several benefits to data scientists on issues such as resource engineering and algorithm testing.
In 2021, I foresee a great increase in the use of artificial intelligence and machine learning in all areas of knowledge, through the concept of hyperautomation, which involves a combination of many technologies, in addition to AI and machine learning, such as augmented analytics, IoT, iBPM and RPA. AI and ML will be increasingly used in the cybersecurity field, in corporate systems, smart cities and smart homes. The AIoT, union of Artificial Intelligence and Internet of Things will also be a key trend, enabling to create solutions with more efficient operations, improving the interaction between machine and humans, and also gain more robustness in data analysis.
Kirk D. Borne, Principal Data Scientist at BoozAllen. Global Speaker. Top #BigData #DataScience #AI Influencer. PhD Astrophysicist.
My top trends in data-related disciplines during 2020 include: the incredible (and incredibly scary) improvements in Automated Narrative Text Generation (such as GPT-3); RPA deployments appearing everywhere; the growth in MLOps tools for the deployment and maintenance of operational enterprise Machine Learning activities; the convergence of AI and DevSecOps in IT operations, as AIOps; a wake-up call to data scientists about the importance of both data drift and concept drift in our deployed models, particularly highlighted in the failure of previously successful models during COVID pandemic; the rise of diverse hybrid data architectures that include edge computing, edge-to-cloud, and fog computing; a rejuvenation of distributed data mining (popular 10-15 years ago) under the label "Federated Machine Learning", enabling incremental model training on distributed data sets without moving the data; the enormous widespread interest and investments in Data Literacy across all organizations and sectors of the market; and the growing awakening to IoT as more than just a monitoring activity (as something you do), but as a technology enabler of a new corporate Observability Strategy (as the reason why you do it).
I expect to see more of these in 2021, especially MLOps, observability, data literacy, and intelligent process automation (AI-enhanced RPA).
Tom Davenport, @tdav Distinguished Professor of IT and Management, Babson College; Senior Advisor, Deloitte Analytics and AI Practice; Digital Fellow, MIT Initiative on the Digital Economy.
- I predicted last year that what is widely now called "MLOps" for ongoing management of machine learning models would become popular. The concept has become popular, but I think actual usage is still high only among large financial services and tech firms.
- I also predicted that translators would become important to data science teams, and many firms do use them, but the role has not become institutionalized
- I predicted "Recognition that the fit of a model to data is only one consideration in whether it is useful or not," and I don't think that has really come to pass. Vendors that emphasize this message have not really prospered.
- The popularity of large scale deep learning models will peak in 2021; data and energy requirements are simply becoming too high for the trend to continue.
- Data science teams will adopt classifications of the roles and skills necessary to succeed with projects; "product manager" will become an important role to ensure deployment.
- External firms will specialize in aspects of "data science as a service," including supplying data for training models and monitoring models over time with MLOps.
Here is a word cloud from these predictions
Carla Gentry, Data Scientist, Analytical Solution
As we look back on another year, the thing that I found most interesting about AI, machine learning and data science is; so many are still talking about python/coding/software, but not many showing real life examples of how these great fields can help society "solve problems" - AI is reduced to chat bots, machine learning becomes character and or object recognition on some "big box site" to classify products or find similar products or recommendations... Now I'm not saying there isn't some really cool stuff being done, especially with Covid being on the forefront of everyone's mind, data science is being used as well as machine learning to find hotspots and/or aid in contact tracing which will help curb the rise in new infections. You know data science, analytics and machine learning were and continue to be used in the aid of all the vaccines that are coming out but as we've seen in the "reliability" and "reproduction" department aka "AstraZeneca/Oxford's COVID vaccine data", data quality and the ability to involve representative samples are still big issues.
For as far as we have come, basic "data 101" issues are still formidable - so let's remember anytime we start a project - check your data, query and follow up on anomalies, don't assume anything. Interesting sidenote, at Thanksgiving dinner, outside and socially distant, my mother shows me this video of two "people dancing" I said ah "that's nice" and smiled - she said - Did you notice those two "people" were robots? Well, dancing is just math, right (wink, however, similar to AI paintings, no soul) 2021 and beyond have limitless capabilities of what we can do, so let's move machine learning and data science forward by remember the basics - reproducibility, transparency, accuracy and is it representative of true life and form - we have to get away from the black box type thinking period - just because it worked in a college setting or lab setting doesn't necessarily mean it's going to work in real life....
Remember this, highways that have on-ramps and off-ramps "short distances" from each other - causing havoc to merge in and out - were designed by engineers that thought it was efficient ... efficient for WHOM? When designing an algorithm or model, KNOW your audience... know your end game, what is the purpose or function and then - of course be logical - but also be PRACTICAL...
Jake Flomenberg is a partner at Wing Venture Capital focusing on AI/ML powered applications who recently published a survey on of the biggest issues facing data science leaders.
2020: COVID-19 was a black swan event that put many models to the test. Unfortunately, most AI does not yet adapt well to rapidly changing situations. As companies were forced to adapt, AI observability and model monitoring became clear needs for anyone working with models in production.
AI for NLP is on the upswing. Human NLP benchmarks are being surpassed. Language processing models like GPT-3 are unlocking new use cases like summarization and code generation. Startups leveraging NLP to unlock value from text continue to proliferate.
AI for healthcare is finally starting to happen - whether it's chemistry, drug discovery proteomics, etc. Over half of all publications in biology involving AI were published in the past 2 years.
2021: Model explainability becomes a mainstream need as data scientists work to create trust across many parties -stakeholders in the organization, regulators, and end users- as well as to better understand their models and offer confidence bounds on performance. Issues ranging from unfair bias, model instability, regulatory requirements, all drive this.
Computer vision proliferates. I expect to see more real world use cases outside of self-driving vehicles as well as more venture funding. I also hope to see computer vision begin to benefit from some of the self-attention strategies in transformer networks that unlocked value in NLP.
Data labeling become more automated. With advances in few shot and transfer learning computers are finally able to step in and reduce the need for humans to label all of the data manually.
IPFConline, Digital Transformation Consulting, Marseille, France.
The year 2020, for the IPFConline team, was marked by 3 important facts in the fields of AI and Data Science:
- the release of OpenAI GPT-3 which is a spectacular advance in Natural Language Processing,
- the real awareness of the limits of Deep Learning in the sense that DL algorithms are too greedy in data, which makes then very difficult to interpret
- the banning of facial recognition when used for mass surveillance in many parts of the world.
Nikita Johnson, Founder, RE.WORK Deep Learning & AI events, #reworkDL #reworkAI #AIforGood
It's been a challenging year, but perhaps, whilst we've been socially distancing, discussions on how we can apply AI for social impact have brought us together, and have been amplified due to the exploration of how we can leverage AI for pandemics, as well as for the greater good. Of course, the recent announcement from DeepMind on their breakthrough in protein folding is another leap for AI for healthcare this year.
I hope we'll see the power of AI in medicine and digital health further advance in 2021, including progress in privacy and personalization, as well as telehealth and mental health applications.
Doug Laney, Innovation Fellow, Data & Analytics Strategy
Throughout 2021, the major industrial focus of applied AI and data science will be to improve the efficiency of life sciences research and healthcare delivery. The Covid-19 crisis has shone a bright spotlight on the gaps and opportunities in healthcare policy, supply chains, operations, diagnoses, prevention, and treatment.
And as individuals continue to search for new ways to work remotely more effectively, intelligent agents will creep into the applications we use to collaborate. They will help us integrate and communicate not just with one another, but with the vast wealth of data, content, and other applications available. Smart devices and assistants will evolve from Q&A oriented to be truly conversational and situationally aware. Moreover, they will learn and mimic our individual ethical standards in dealing with us and others.
Additionally, we'll see AI start to manage data itself. Data volume and velocity have always just been a matter of infrastructure scaling. However, the growing and ever-evolving variety of data to be integrated, cleansed, organized, stored, and accessed for transactional and analytic purposes has long been a monumental manual architectural and engineering endeavor. Self-organizing data management capabilities will emerge to ease this burden.
Ronald van Loon, Helping data driven companies generating value; Top10Influencer in #AI, #BigData, #DataScience, #IoT, #MachineLearning, #Analytics, #Cloud, #5G
In 2020, Conversational AI accelerated at an exceptional rate as businesses responded to the challenges brought on by the COVID-19 pandemic. Urgent demands for new digital services, changing customer needs, and the shift to remote work drove strategic adoption of this technology at scale. The rise of AI APIs, such as Conversational AI APIs and Document AI APIs, made it easier and faster to promote AI accessibility and eliminated the need for specialized AI skills or experience.
For 2021, embedded AI will be a top trend as companies focus on increasing digital transformation momentum and quickly innovating with AI via intuitive, simplified AI capabilities for business applications. This will ensure that the benefits and value of AI are rapidly realized while allowing organizations to rapidly modernize their business. Similarly, machine learning applications for business users will drive democratization of AI in the enterprise through streamlined implementation and processes, automated workflows, and faster insight generation.
Bill Schmarzo, @schmarzo, Dean of Big Data | Recognized innovator, educator, practitioner Data Science, Design Thinking, Data Monetization
Main 2020 Developments:
- Leveraging AI to create autonomous assets that appreciate, not depreciate, in value the more that they are used becomes a reality (thanks Elon Musk)
- Orphaned Analytics - ML models built for one-off analytic purposes but never engineered for re-usability or continuous learning and adapting - continue to be an organizational disappointment
- Continued advancements in open source tools for development, operationalization and management of AI/ML models
- Under-appreciating the unintended consequences that result from bad estimates of the costs of False Positive and False Negative errors
- While Data Monetization continues as CIO's #1 challenge, many lack a business-centric value engineering methodology to identify and prioritize where and how AI/ML can derive new sources of customer, product and operational value
- Organizations begin to appreciate the economic value of data and analytics, digital assets that never wear out and get more valuable the more that they are used
- Autonomous Analytics start to replace Orphaned Analytics in leading organizations
Kate Strachnyi, Founder of Story by Data & DATAcated Academy
In 2020, we witnessed a shift away from static dashboards in the direction of interactive data stories that are developed automatically and delivery key insights to stakeholders. The data messages are contextualized and tailored to the audience (people see the insights that matter to them). We've also seen software that allows business users to 'ask' questions by typing them into a search bar; allowing them to get closer to the data and reduce the effort and time it takes to obtain data-driven answers.
In 2021, we can expect to see a push to increase data literacy skills among professionals in all industries and roles (not just the data experts). There will be a move towards educating employees at all levels on how to read, analyze, interpret, and communicate with data. Powered by innovative technology, we will spend less time building dashboards and more time focusing on making data-informed decisions.
Dr Mark van Rijmenam, VanRijmenam, thinking about technology and the impact on business & society. Author of 3 mgt books. Founder of Datafloq.com. Fighting fake news, bots and trolls with Mavin.org.
In 2020, we saw the arrival of advanced language models with billions of parameters that were able to wow people with their accuracy. An article in The Guardian even showed us how advanced this model is.
However, in 2021 we will likely see the first language model with a trillion parameters taking the technology to another level. Although these models will likely first remain in the research domain, tech such as GPT-3 will find its way to businesses and start interacting directly with customers. Another development we will see in 2021 is the increased attention and requirements for ethical AI. The more AI becomes commonplace in society, the more citizens will require an ethical approach to AI. In 2020, we saw Amsterdam and Helsinki launching algorithm registries to bring transparency to public deployments of AI and we will see this more often in 2021.
- AI, Analytics, Machine Learning, Data Science, Deep Learning Research Main Developments in 2020 and Key Trends for 2021
- AI, Analytics, Machine Learning, Data Science, Deep Learning Research Main Developments in 2019 and Key Trends for 2020
- AI, Analytics, Machine Learning, Data Science, Deep Learning Technology Main Developments in 2019 and Key Trends for 2020