What will applied AI look like in 2022?

Hear from the CIO, CTO and other C-level and senior executives on data and AI strategies at the Future of Work Summit on January 12, 2022. Learn more


AI adoption has increased in the last 18 months. In addition to Joe McCendrick, the founding author of HBR, professionals working on AI will easily validate this statement. It seems that Google search is not very mysterious: when prompted with “AI adoption”, its auto-complete “skyrocketed in the last 18 months” comes out.

Both the hypothetical evidence and the surveys we know point in the same direction. In the case: AI adoption in the Enterprise 2021 survey by O’Reilly in early 2021 has three times more responses than in 2020, and company culture is no longer the most important barrier to adoption.

In other words, more people are working with AI, it is now being taken seriously, and maturity is increasing. That’s all good news. This means that AI is no longer a game that researchers are playing – it is being implemented, which is central to the likes of Microsoft and Amazon and beyond.

Check out the columns below that we expect to be implemented in 2022.

AI chips

In general, when discussing AI, people think about models and data – and for good reason. These are the parts that most practitioners think they can bring a little control over, while the hardware is largely invisible and its capabilities are seen as fixed. But is that the case?

The so-called AI chips, a new generation of hardware designed to optimize AI-related workloads, are seeing explosive growth and innovation. Cloud mainstays like Google and Amazon are building new AI chips for their datacenters – TPU and Trainium, respectively. Nvidia dominates this market and has built an empire around its hardware and software ecosystem.

Intel is considering a hold, whether through acquisitions or its own R&D. Arm’s position remains somewhat unclear, with acquisitions announced by Nvidia facing regulatory scrutiny. In addition, we have many different new players on their adoption journey, some of whom – such as Graphcore and Sambanova – have already reached the Unicorn status.

For applicable AI this means that choosing where to run the AI ​​workload is not just a matter of deciding between Intel CPUs and Nvidia GPUs. There are many parameters to consider now, and that development is important not only for machine learning engineers, but also for AI practitioners and users. The AI ​​workload runs more economically and efficiently which means that the market will have more resources to use elsewhere with faster time.

MLOps and data centricity

Choosing which hardware to run on the AI ​​workload can be considered as part of an end-to-end process of AI model development and deployment, called MLOps – the art and science of bringing machine learning into production. To draw a connection with AI chips, standards and projects such as ONNX and Apache TVM can help eliminate the tedious process of machine learning model deployment on various targets.

In 2021, with the lessons learned from implementing AI, the emphasis is now shifting from the glossy new model to perhaps more physical, but practical, aspects such as data quality and data pipeline management, which are important components of all MLOps. Like any discipline, MLOps sees many products in the market, each focusing on different aspects.

Some products focus more on data, others on data pipelines and some cover both. Some products monitor and inspect items such as inputs and outputs for model, drift, loss, accuracy and recall accuracy for data. Others do the same, yet different things around data pipelines.

Data-centric products meet the needs of data scientists and data science leads and perhaps machine learning engineers and data analysts. Data pipeline-centric products are more oriented towards DataOps engineers.

In 2021, people tried to name various events related to MLOps, slicing and dice MLOps domain, implement data version control and continuous machine learning and implement test-based development equivalents for data, among other things.

However, what we see as the most profound shift is the emphasis on so-called data-centric AI. Well known AI thought leaders and practitioners such as Andrew Ng and Chris Ray have discussed this concept, which is surprisingly simple in its origins.

We have now reached a stage where machine learning models are sufficiently developed and work well in practice. So much so, in fact, that it doesn’t make much sense to focus on developing new models or on fine-tuning to perfection. According to the data-centric view, what AI practitioners should do instead is focus on their data: cleaning, refining, validating and enriching data can go a long way towards improving the results of AI projects.

Large language models, multimodal models and hybrid AI

Large language models (LLMs) may not be the first thing to consider when discussing applied AI. However, those who know believe that LLM can internalize the basic forms of language, be it biology, chemistry or human language, and we are seeing an increasing number of unusual applications of LLM.

To support those claims, it is worth noting that we are already seeing a kind of ecosystem being built around LLM, with the GPT-3 API being commercially available through OpenAI in conjunction with Microsoft. This ecosystem consists mostly of companies offering copywriting services such as marketing copy, email and LinkedIn messages. They haven’t set the market on fire yet, but that’s just the beginning.

We anticipate that LLM will adopt many new ways in 2022 and lead to innovative products: through more options for customization of LLMs such as GPT-3; Through more LLM creation options, such as Nvidia’s NeMo Megatron; And through LLMs-a-a-service offering, such as from SambaNova.

As VentureBit’s own Kyle Wiggers noted in a recent episode, multimodal models are fast becoming a reality. This year, OpenAI introduced two multimodal models, DALL-E and CLIP, which research labs claim is “a step towards the system.” [a] A deeper understanding of the world. “If LLM has anything to look forward to, we can reasonably expect to see commercial applications of multimodal models in 2022.

Another important direction is hybrid AI, which is about disseminating knowledge in machine learning. Leading figures such as Intel’s Car Singer, LinkedIn’s Mike Dillinger and Hybrid Intelligence Center’s Frank Van Harmelan point to the importance of knowledge organization in the form of knowledge graphs for the future of AI. It remains to be seen whether the hybrid AI will produce the AI ​​application implemented in 2022.

Applied AI in healthcare and manufacturing

Let’s get some more support: Promising domains for AI implemented in 2022. The Enterprise 2021 survey cites technology and financial services as the two leading domains in adopting AI in O’Reilly’s AI adoption. This is hardly surprising given the technology industry’s desire to “eat its own dog food” and the financial industry’s desire to gain every inch of competitive advantage possible using its deep pockets.

But what happens beyond those two industries? O’Reilly’s survey cites health care as the third domain to adopt AI, and this is consistent with our own experience. As state of AI authors Nathan Benaich and Ian Hogarth noted in 2020, biology and healthcare are seeing their AI moment. This wave of adoption was already in motion, and it gained momentum with the advent of COVID-19.

“For example, ‘I think this gene is responsible for this disease, let’s take action against it and find out if it’s true.’ Then there are the more software-driven ones in this new age of pharma. They mostly look at large-scale experiments, and they ask many questions at once. In fairness, they let the data map what they should focus on, ” Said summarizing the AI-driven approach.

The only way to validate whether the new-age pharma approach works is to produce drug candidates who actually prove to be useful in the clinic and eventually get those drugs approved, Benaich added. Of those “new age pharma” companies, Recursion Pharmaceuticals filed for IPO in April 2021 and Exscientia for IPO in September 2021. They both have assets generated by their machine learning-based approach that are actually being used medically.

When it comes to manufacturing, there are several reasons why we choose to highlight AI out of the many domains behind adopting AI. First of all, he suffers from a labor shortage which is the kind of AI that can help him overcome it. As many as 2.1 million manufacturing jobs could remain unfinished by 2030, according to a study published by Deloitte and The Manufacturing Institute. AI solutions that perform functions such as automated physical product inspection fall into this category.

Second, the nature of the industrial application requires a very precise combination of data with the physical world. This, as some have noted, lends itself well to hybrid AI approaches.

And last but not least, hard data. According to a 2021 survey by The Manufacturer, 65% of manufacturing sector leaders are working to pilot AI. Implementation in the warehouse alone is expected to hit a compound annual growth rate of 57.2% over the next five years.

Venturebeat

VentureBeat’s mission is to become a digital town square for technical decision makers to gain knowledge about transformative technology and practices. Our site delivers essential information on data technologies and strategies so you can lead your organizations. We invite you to access, to become a member of our community:

  • Up-to-date information on topics of interest to you
  • Our newsletters
  • Gated idea-leader content and discounted access to our precious events, such as Transform 2021: Learn more
  • Networking features and more

Become a member

Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *