ARTIFICIAL INTELLIGENCE (AI) | From large corporations to startups to independent research labs, machine learning teams work hard every day to build and deploy models that improve everyday life. As the end of the year approaches, it’s good to stop and see their progress.
In many ways, the year 2022 begins and ends with generative AI. Last year was launched Dall-E 2 by OpenAI (April), but also by Midjourney (April), from Stability Diffusion 2 from Stability AI (August) and from ChatGPT (November). Technological breakthroughs that happen almost every week, whether transforming robotics or improving the study of the genome, make fewer headlines but are just as revolutionary.
For most machine learning specialists, the problem today is not the lack of research, tools or new techniques, but the fact that they cannot keep up with all these advances!
Before we look to the future, here is a review of the past year.
- AI equity will decline before improving: TRUE. Overall, the impact of AI behaving in a discriminatory way is probably not fully known, but this year leaves a lot to be desired. For example, it is now clear that harmful biases lurk in large text-to-image conversion models and language models. Model bias also continues to appear in more common models, affecting everything from health outcomes to claims. Finally, there is still a long way to go to ensure better diversity in AI recruitment and ethics.
- Companies blindly stop implementing artificial intelligence: PARTIALLY TRUE. As the adoption of machine learning monitoring accelerates and market leaders emerge, the reality is that many teams still do not have a monitoring plan in place to detect and diagnose problems early. This is especially true for teams that have implemented computer vision and natural language processing models, as the tools to monitor things like integration operation are still very new.
- The highlight of the citizen researcher: PARTIALLY TRUE. While the adoption of low-code tools remains a factor in the democratization of data science, it is somehow overshadowed by the revolution emerging around large language models with text as the universal interface.
- The machine learning infrastructure ecosystem will become denser and more complex: TRUE. With investment in AI and machine learning infrastructure tools increasing this year, the ecosystem is only more crowded. Overall, 85.7% of big data experts and machine learning engineers say they still sometimes “struggle to navigate a confusing and messy machine learning infrastructure ecosystem.”
- The number of machine learning engineering positions will exceed the number of available talent, creating a talent shortage: TRUE. While recent layoffs are a concern, the past year has been particularly marked by labor shortages across the economy, particularly in data science and machine learning.
In 2023, here are the AI trends to see.
1. Generative AI will go mainstream (so will its growing pains)
Generative AI captures the public’s imagination in a way that few technological advances have done since the advent of cinema more than 100 years ago. With powerful apps like Github, co-pilot Where ChatGPT, which has already proven itself, many companies are eager to adopt this technology on a larger scale. However, generative AI remains uncharted territory. There will be many elements to be clarified in 2023, especially regarding bias, copyrights, scalability, security and how to monitor this new technology. In summary, generative AI will require a network, and you will have to build that network.
2. Economic uncertainty will be a major driver for the machine learning infrastructure market
AI is likely to become more important as inflation and economic disruption put pressure on businesses to become more efficient and productive. Given shifting priorities, this is the end of the days when core machine learning teams took months or even years to build and maintain feature stores or internal monitoring tools. . Buying rather than building is likely to become more common, not least because teams must prioritize projects that increase the company’s revenue in the short term. Given the economic environment, it is not unlikely that procurement pressures, or even layoffs, will affect machine learning teams in some industries. In this context, only the tools MLOps the strongest who bring real added value to the teams will thrive. Expect tools like orchestration platforms to reveal some outdated assumptions about connecting many disparate machine learning tools, and leaders outside of the category scrambling to raise money or going out of business.
3. The most successful platforms will attack traditional players
That’s what happened to DevOps and this is what is happening now with MLOps: in technical areas, the best platforms tend to win. Given the complexity of modern machine learning, machine learning teams need deeper tools at each stage of the model lifecycle. As a result, platforms that offer integrated, so-called end-to-end solutions that emerged a decade ago to support the work of big data experts and machine learning teams are losing their share of developers and facing layoffs. Even big players like Amazon (with SageMaker) and Google (with Vertex) currently do not reflect the technical depth needed for each part of the machine learning lifecycle, although a wave of consolidation may change that.
4. Working with unstructured data will no longer be optional
Unstructured data is everywhere. According to several estimates, 80% of the data generated in the world is in the form of unstructured images, text, video or audio files. In recent years, some of the most powerful modern applications of machine learning have leveraged unstructured data. Any machine learning platform not designed to handle unstructured use cases may not be relevant or have limited growth prospects. At the same time, machine learning teams that find ways to leverage computer vision or natural language processing models, even if it’s just applying a pre-trained model to a case restricted from commercial use, can find new competitive advantages.
FinallyWhile not all prospects seem bright or positive, there are plenty of reasons to be optimistic about the future of AI and machine learning teams. It only remains to hope that everyone can start the year 2023 with the thoughts of making this sector better!
Article translated from Forbes US – Author: Aparna Dinakaran
<< For at læse Souralso: AI til gavn for virksomhedernes bæredygtighed >>>