Analyze – building and scaling models with trust and transparency
Enterprises are either building AI or buying AI solutions to address specific requirements. In the case of a build scenario, companies would benefit significantly from commercially available data science tools such as Watson Studio. IBM's Watson Studio not only allows you to make significant productivity gains but also ensures collaboration among the different data scientists and user personas.
Investing in building AI and retraining employees can have a significant payoff. Pioneers across multiple industries are building AI and separating themselves from laggards:
- In construction, they're using AI to optimize infrastructure design and customization.
- In healthcare, companies are using AI to predict health problems and disease symptoms.
- In life science, organizations are advancing image analysis to research drug effects.
- In financial services, companies are using AI to assist in fraud analysis and investigation.
- Finally, autonomous vehicles are using AI to adapt to changing conditions in vehicles, while call centers are using AI for automating customer service.
However, several hurdles remain, and enterprises face significant challenges in operationalizing AI value.
There are three areas that we need to tackle:
- Data: 80% of time is spent preparing data versus building AI models.
- Talent: 65% find it difficult to fund or acquire AI skills.
- Trust: 44% say it's very challenging to build trust in AI outcomes.
Source: 2019 Forrester, Challenges That Hold Firms Back From Achieving AI Aspirations.
Also, it's worth pointing out that building AI models is the easy part. The real challenge lies in deploying those AI models into production, monitoring them for accuracy and drift detection, and ensuring that this becomes the norm.
IBM's AI tools and runtimes on Cloud Pak for Data present a differentiated and extremely strong set of capabilities. Supported by the Red Hat OpenShift and Cloud Pak for Data strategy, IBM is in a position to set and lead the market for AI tools. There are plenty of point AI solutions from niche vendors in the market, as evidenced from the numerous analyst reports; however, none of them are solving the problem of putting AI into production in a satisfactory manner. The differentiation that IBM brings to the market is the full end-to-end AI life cycle:
Figure 1.5 – AI life cycle
Customers are looking for an integrated platform for a few reasons. Before we get to these reasons, the following teams care about the integrated platform:
- Data science teams are looking for integrated systems to manage assets across the AI life cycle and across project team members.
- Chief Data Officer are looking to govern AI models and the data associated with them. Chief Risk Officer (CRO) are looking to control the risks that these models expose by being integrated with business processes.
- Extended AI application teams need integration so that they can build, deploy, and run seamlessly. In some situations, Chief information officer (CIOs)/business technology teams who want to de-risk and reduce the costs of taking an AI application to production are responsible for delivering a platform.
Customer Use Case
A Fortune 500 US bank is looking for a solution in order to rapidly deploy machine learning projects to production. The first step in this effort is to put in place a mechanism that allows project teams to deliver pilots without having to go through full risk management processes (from corporate risk/MRM teams). They call this a soft launch, which will work with some production data. The timeline to roll out projects is 6-9 months from conceptualization to pilot completion. This requirement is being championed (and will need to be delivered across the bank) by the business technology team (who are responsible for the AI operations portal). The idea is that this will take the load away from MRM folks who have too much on their plate but still have a clear view of how and what risk was evaluated. LOB will be using the solution every week to retrain models. However, before that, they will upload a CSV file, check any real-time responses, and pump data to verify that the model is meeting strategy goals. All this must be auto-documented.
One of the key differentiators for IBM's AI life cycle is AutoAI, which allows data scientists to create multiple AI models and score them for accuracy. Some of these tests are not supposed to be black and white.
Several customers are beginning to automate AI development. Due to this, the following question arises: why automate model development? Because if you can automate the AI life cycle, you can enhance your success rate.
An automated AI life cycle allows you to do the following:
- Expand your talent pool: This lowers the skills required to build and operationalize AI models
- Speed up time to delivery: This is done by minimizing mundane tasks.
- Increase the readiness of AI-powered apps: This is done by optimizing model accuracy and KPIs.
- Deliver real-time governance: This improves trust and transparency by ensuring model management, governance, explainability, and versioning.
Next, we will explore how AI is operationalized in enterprises to address specific use cases and drive business value.
