1. Responsible AI Goes From Aspiration to a Foundational Requirement
In 2021, the AI industry had an all-talk and no-walk problem. While you can read dozens of think pieces and thought leadership articles about responsible AI in 2021 (including our own World Economic Forum Agenda blog post), adoption of responsible AI principles was low. According to our Appen 2021 State of AI report, concern for AI ethics was at just 41% for technologists and 33% for business leaders. In 2022, the stakes get higher and businesses will begin to recognize that responsible AI leads to better business outcomes. The business leaders will catch up to the technologists in understanding the importance of responsible AI. And, they’ll begin to see how the up-front investment will pay off for their business. When responsible AI principles are properly implemented, they protect a business’s brand and ensure that the AI project works as expected. Entering 2022, we also have a well-established and thoroughly reviewed set of responsible AI principles. They include:- Unbiased data
- Fair treatment of data collectors and labelers
- The need for AI projects to promote social good and prevent social harm
2. Data for AI Lifecycle Becomes Critical for AI Programs
Recent statistics and trends show that AI programs are maturing and AI is increasingly present everywhere. AI is powering business operations and shaping product development. According to the Appen 2021 State of AI report, AI budgets have increased in the last year. This shows a recognition by business leaders that they must invest in AI to ensure success. One of the key takeaways from 2021 is that businesses, even those with mature AI data science sectors, are struggling with data. What businesses are realizing is the vastness of the amount of data that’s needed for AI model development, training, and re-training. Because so much data is needed for a successful AI lifecycle, many businesses are choosing to partner with external training data providers to deploy and update AI projects at scale. The fact that a majority of organizations are pairing with external data partners shows the challenge of continuous data sourcing, preparation, evaluation, and production. AI projects need more data and faster than ever before. This can only be achieved through automation, especially around data sourcing and preparation. This need for data will shift in 2022. Companies will still need just as much data, but a new discipline will be developed. Data for the AI lifecycle will focus on the development of tools and best practices that enable businesses to manage the entire AI lifecycle, from data acquisition to data versioning and all the way to model retraining.3. Rise of Synthetic Data
As more and more data is needed to satisfy data-hungry AI programs and model retraining, the industry is going to see new ways for businesses to acquire data. While the only solution for more data at the speed needed by these companies is an external data partner, another solution is on the horizon. Generative AI can create synthetic data, which can be used to train AI models. While currently only accounting for 1% of the data on the market today, Gartner believes that generative AI will account for 10% of all data produced by 2025. Currently, generative AI is being used to address key challenges such as generating 3D worlds for AR/VR and for training autonomous vehicles. Gartner also forecasts that by 2024, the use of synthetic data will halve the volume of real data needed for machine learning. The use of synthetic data complements and accelerates the data acquisition process because it needs less processing, security, and labeling than real-world data which is subject to responsible AI principles. In 2022, you can expect a lot more businesses and machine learning models using and experimenting with synthetic data. Generative AI models can learn from themselves and generate new data, which is cost-effective and improves efficiency for businesses. With these benefits, it’s obvious why many businesses are excited about generative AI and synthetic data. And, as more companies experiment with and implement synthetic data and generative AI, we’ll see new use cases developed over the next few years.4. Acceleration of Internal Efficiency Use Cases
Some great news for the industry: AI budgets are on the rise, according to the Appen 2021 State of AI report. 74% of respondents reported that they have AI budgets of over $500k. As well, 67% of business leaders say that their AI projects have “shown meaningful ROI.” As budgets grow and the variety of use cases expands, it’s not surprising the number one most popular use case, at 62%, is supporting internal operations. The next most common use cases follow a similar pattern, businesses are using AI to make their own internal operations more efficient with:- 55% looking to improve their understanding of corporate data
- 54% looking to improve productivity and efficiency of internal business processes
- They will need to focus more attention on deploying platforms that enable them to eliminate data silos and centrally manage data
- They will need to work internally or with partners to develop strategies focused on being able to manage data throughout the entire AI lifecycle.
5. Model Evaluation and Tuning Becomes Mainstream
A realization has begun to slowly reverberate through the AI technology community: building an AI machine learning model isn’t just one-and-done. The model needs regular evaluation, tuning, and retraining. In 2022, this awareness will become common knowledge. Machine learning models are dynamic, they can’t just be deployed and left to their own devices. Just like a car that needs its alignment adjusted regularly, machine learning models can develop drift over time. This drift can make the machine learning model results less and less accurate over time. Machine learning models must be reviewed and updated based on its ongoing results and any changes to infrastructure, data sources, and business models. According to our report, the knowledge that machine learning models must be regularly reviewed and updated took a huge leap in 2021. We found:- 87% of organizations update their models at least quarterly, up from 80% last year
- 57% update their models at least monthly
- 91% of large organizations update their models at least quarterly
- Organizations that use external data providers are most likely to update their models at least monthly.
