Today, many companies are making plans to adopt AI technologies to gain or maintain competitive advantage. In 2021, Gartner found strong indicators for increased industry wide funding in AI. AI adoption continues apace, but while many companies adopt AI, their ability to leverage value from AI nevertheless remains isolated or fragmented.
What if everyone in your business could get the benefits of AI?
In this post we will take a look at AI democratisation and its benefits, the challenges it poses and how to overcome them, and a few examples.
The ability to democratise AI fully will be key in making the most of the AI revolution currently under way.
What is AI democratisation?
Simply put, AI democratisation is the availability and accessibility of AI to non-specialist users.
Imagine the new sales director, rather than wrestling with a spreadsheet, can run their data through a powerful and trusted, easily discoverable sales model. They are instantly empowered.
We are already seeing AI democratisation happening in the consumer space. From entertainment apps to serious use cases in the medical and legal domains, users can access a variety of AI models and add their own custom data to experiment with outcomes. They can do this on their own initiative, without significant upfront investment. Datasets, algorithms, and models alike are easily discoverable, often from a simple search engine query, and there are usually no upfront costs.
Examples include API marketplaces like RapidAPI, for discovering new datasets, and AI vendors like EndlessMedical and OpenAI that provide access to powerful models. In some cases, like OpenAI’s GPT-3, the user can augment the model with their own data.
Developers have even more options available to them, with open source offerings that range from those curated and hosted by specialist communities like Huggingface, to broad developer communities hosted on Github or Kaggle.
With such a range of choice the public consumer space is currently well ahead of what most organisations are able to offer their own internal users.
The Big 3 cloud vendors understand organisations’ growing need to build and run AI models. As a result they offer comprehensive machine learning platform solutions. AWS has the Sagemaker platform, Azure offers Azure Machine Learning, and GCP has Vertex AI. They also offer analytics tools and make it easy to connect them to data sources. AWS Quicksight can be connected to data sources like Redshift, Aurora, RDS, and S3 to produce machine learning forecasting or anomaly detection insights, whereas Azure and GCP have similar offerings through PowerBI, Tableau, and Looker.
Lowering the barrier to entry for business data analysts even further is Amazon SageMaker Canvas, a no-code offering, which has a visual point-and-click interface. If the use case fits, an organisation can even use one of the available off-the-shelf APIs for vision AI, text-to-speech and speech-to-text, natural language AI, and more.
These services do not stop with the Big 3. Other data, analytics, and AI vendors like Databricks and Snowflake are partnering with them to offer their own products in the cloud marketplace.
Although cloud vendors are making it easier to build and deploy AI solutions, there is still a tendency to build predictive models in a data science silo. Such an approach does not scale or integrate well with the rest of the business. Instead, cross-functional teams need to work together and include domain experts to build accessible productionised solutions.
However it is not only about collaboration. AI democratisation also reflects the decision-making culture of an organisation. Rigid, hierarchical organisations may be more reluctant to empower their employees through AI democratisation, yet they may also be the ones who stand the most to gain, due to their rich experience and access to large amounts of historical data. This is how Conway’s Law will have an influence on individuals’ ability to access AI in the future.
Supposing an organisation sees the value in making AI accessible, it still needs to be approached with great care. A haphazard approach to AI democratisation can quickly lead to divergence across the enterprise, resulting in conflicting “insights” and “predictions” that cannot be trusted. In such a scenario decisions can no longer be made with confidence, and will ultimately slow or even reverse adoption of AI in the business.
We want to build an AI product that is of high quality so that non-specialist consumers outside of the domain can use it with confidence. Fortunately, there are guidelines that will increase the likelihood of building trustworthy AI.
The data mesh framework advises us to approach data with a product mindset and we can approach AI in the same way. In the first instance, the AI should be easily discoverable by the user, for example using a common search tool. Once discovered, the user should be able to assess how useful the AI is to their use case using the documentation provided. There should be clear quality indicators that are measurable against specific use cases. The documentation should explain what the AI does, how to use it, datasets used during training, benchmark results, and so on. This information is not ad hoc, but forms explicit contracts.
Another key component of a trustworthy AI is explainability. Simply put this refers to how understandable and interpretable a model and its predictions are. This is an active area of research and there is broad agreement on certain approaches, such as preferring glass box models over black box models. It ensures that any decisions based on predictions provided by the AI can be justified with confidence. For example indicating if a model has bias, or what input features have the most influence on the outcome.
Let’s look at a couple of examples of AI democratisation.
When a retailer markets its goods it keeps a close eye on changes in the market to optimise its marketing campaigns. Traditionally, someone would trawl relevant media outlets, social media, etc. Much of this is automatable now, but it remains difficult to interpret and understand all the available data. A trusted AI that can perform analysis (e.g. sentiment analysis) adds a further layer of interpretation to this process, which can enhance decision-making. The marketer’s ability to find and access a suitable AI when needed, and feed large amounts of data to it, renders decision-making much more flexible.
Or consider a large events company that sends out teams to run events. Each event requires staff with a variety of skills, such as front-of-house, waiting, cooking, DJ-ing, sound technician, etc. Every staff member has a certain skill profile, familiarity with other staff members, proximity to the event, availability, and so on. An event manager would want to run multiple combinations and scenarios, and having discoverable access to both the relevant data as well as a trusted AI that can inform different scenarios will help them pick the best team for the event.
In each of these cases AI democratisation allows the decision maker to augment their understanding without first having to overcome organisational barriers. They may have to put in a request to use the AI, for governance purposes, but it will still be discoverable in the first instance, and usable once access is approved and granted.
Businesses work in uncertain environments. Decision makers at all levels of the business rely on having access to the right data and tools to gain insights and take action. However, there are both technical and organisational barriers to accessing the right data and tools for the task, and the result is more often than not to use whatever is most convenient. It is no wonder that spreadsheets and client side databases have been tools of choice for so many years.
Data mesh promotes data democratisation, setting the stage for consumers in the business to access and leverage data quickly in order to gain actionable insights through advanced capabilities like machine learning. Similarly, AI democratisation enables non-specialist users to adopt and gain value from AI tooling and practices in the organisation.
In short, it makes AI more convenient and thus vastly more accessible.
With legacy approaches to organisational AI the technology takes centre stage (e.g. RedShift ML, BigQuery ML etc.). The mesh approach promotes a move away from a technology centred approach to one concerned with business value and convenience to the consumer.
Some of the benefits of AI democratisation in an organisation include:
- Reducing barriers to entry due to ease of use
- Reducing the time required to upskill in AI
- Increasing the speed of AI adoption
- Reducing the overall cost of AI
Data democratisation enhances business operations. By provisioning a data-as-a-service architecture, the data mesh brings agility to business operations. By extending this to AI-as-a-service, business operations extend AI capabilities closer to the decision makers. As better practices are adopted, business intelligence is updated more quickly, and the business can be more responsive to changes in the market.
Businesses need to make decisions every day to succeed in a risky market environment. AI democratisation brings the power of AI closer to the decision makers, accelerating their access to business insights that reduce the risk and optimise their chances of success.
Interested in seeing our latest blogs as soon as they get released? Sign up for our newsletter using the form below, and also follow us on LinkedIn.