Four predictions for AI in 2023

From large corporations to start-ups to independent research labs, machine learning (ML) teams are hard at work every day to create and implement models that improve the way we live and work. As the end of the year approaches, it’s worth pausing to marvel at your progress.

In many ways, that story in 2022 begins and ends with generative AI. Last year, not only OpenAI’s Dall-E 2 (April), but also Midjourney (April), Stability AI’s Stable Diffusion 2 (August), and ChatGPT (November) were released. Less headline-grabbing, but no less transformative, are the steady near-weekly technical advances in everything from robotics transformers a improved genome studies.

For most technical ML professionals, the current problem is not a lack of new research, tools, or techniques, but rather that they can’t keep up!

Before we look ahead, here’s a report card on last year predictions.

  • AI fairness will get worse before it gets better: CERTAIN. The full impact of AI acting in discriminatory ways may not be fully known, but this year leaves a lot to be desired. For example, it is now clear that harmful biases Watch in prominent text-to-image and large language models. Model bias continues to rear its ugly head in more everyday models, impacting everything from health outcomes a insurance claims. Finally, there is still a A long way to go to ensure greater diversity in recruitment and AI ethics.
  • Companies will stop sending AI Blind: PARTIAL CREDIT. While the adoption of ML observability is accelerating and market leaders are emerging, the reality is that many teams I have not yet set up monitoring to quickly detect and diagnose problems with models in production. That’s particularly true for teams with implemented computer vision and natural language processing models, since tools for monitoring things like embedding drift they are still so new.
  • The citizen data scientist will rise up: PARTIAL CREDIT. While the adoption of low-code tools continues to be a factor in the democratization of data science, it is somewhat overshadowed by the revolution emerging around large language models with text as universal interface.
  • The ML infrastructure ecosystem will become more crowded and complex: CERTAIN. With investment in AI and machine learning infrastructure tools growing this year, the space is only more crowded. In total, 85.7% of data scientists and ML engineers say sometimes they still have “trouble navigating a confusing/crowded ML infrastructure space”.
  • ML engineering jobs will outpace available talent, creating a talent crunch: CERTAIN. While the recent layoffs are cause for concern, the past year is mostly a story of labor shortage throughout the economy especially in data science and machine learning engineering.

To cap off the year, here are some things to watch out for in 2023.

1. Generative AI will go mainstream (and so will its growth issues)

Generative AI is capturing the public imagination in a way that few technical advances have since the beginning. advent from the movie over 100 years ago. With powerful applications like Github Copilot and ChatGPT already proving valuable, many companies are eager to adopt the technology more widely. However, generative AI is still a wild west. There’s a lot to unpack over the course of the next year prejudice, Copyright, scalability, safetyand how to monitor this new technology. In short, the generative AI will take a village, and we need to build that village.

2. Economic uncertainty will be a melting pot for the ML infrastructure market

AI is likely to become increasingly important as inflation and economic turmoil put pressure on businesses to deliver greater efficiency and productivity. Given shifting priorities, the days of core ML teams spending months or years building and maintaining proprietary feature stores or internal monitoring tools are likely numbered. Buy rather than build is likely to become more common, particularly as teams must prioritize projects that move the revenue needle in the near term. Given the economic environment, it is also not unlikely that acquisition pressure and even layoffs could affect LA teams in certain sectors. In that context, only the strongest MLOps tools that add real value for teams will thrive. Expect tools like orchestration platforms, which reflect outdated assumptions about connecting many disparate ML tools, alongside non-category leaders to fight to raise capital or shut down.

3. Best-in-class rigs will weed out legacy players

It happened in DevOps and now it’s happening in MLOps: In technical fields, the best platforms tend to win the day. Given the complexity of modern machine learning, ML teams are demanding more depth from the tools at each stage of the model lifecycle. As a result, end-to-end platforms that emerged a decade ago to empower both citizen data scientists and ML teams are losing development share and suffering layoffs. Even the big players like Amazon (SageMaker) and Google (Vertex) that offer end-to-end solutions currently don’t reflect the technical depth needed for each part of the ML lifecycle, although a wave of consolidation could change that.

4. Working with unstructured data will no longer be optional

Unstructured data is everywhere. According to multiple estimates, 80% of the data generated globally is in the form of unstructured images, text, video or audio. In recent years, some of the most powerful modern machine learning applications, from large language models like ChatGPT to computer vision models that can detect cancer or rare medical disorders, take advantage of unstructured data. Any ML platform that is not designed to handle unstructured use cases risks being irrelevant or having limited growth prospects. At the same time, ML teams that find ways to take advantage of computer vision or NLP models, even if they only apply a pre-trained model to a limited business use case, may find new competitive advantages.

Conclution

While all the prospects here may not seem bright and sunny, there’s a lot to be optimistic about when it comes to the future of AI and ML teams. We hope everyone can start 2023 with their sights set on making this industry better!

Leave a Reply

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