Strategy Series: AI 2020 (Part 1)
In this Article I want to write few guiding points towards how to define AI Strategy for 2020. I am going to use some of the reference from Gartner reports and other industry bench to prepare ensemble of the trends. Key guiding trends of 2020 would be as below.
How to measure the progress: while the below strategy guidelines listed seems very obvious for many practitioners, and many say all organizations are implementing the below levers, the devil is always in details. The key question everyone should ask is how anyone can measure maturity level in particular lever. For this purpose we created 4 stages of Maturity levels
Maturity Level 1: Experimenting: Essentially where organizations have done few POCs on this lever and haven’t implemented anything in production yet
Maturity Level 2: Incubating: Where organization is developing any projects on this, or have implemented few projects which involve limited size of data, but haven’t integrated at extensive data sources
Maturity Level 3: Leading at scale: This is the maturity level where organizations implemented this at scale and using extensive data sources leveraging various technology landscape like Big Data, cloud etc and also extended to multiple lines of business rather than at Silos, this is the saturation level for organizations with respect to business benefits
Maturity Level 4: Leading the Industry: This stage is where organizations actually lead the innovation and research on par with technology companies. This might be debated whether there is enough business benefits for the organization to be at this stage. But this provides an indicator based on their competitor landscape
Now coming to strategy levers:
1) Augmented Analytics: Augments analytics will be key trend in defining AI strategy and path forward for organizations. Augmented analytics automates finding and surfacing the most important insights or changes in the business to optimize decision making. It is very important to remove user bias / data scientist bias in defining features.
Example: Suppose there is a feature defined in model which targets at millennials, how exactly do the data scientists know about millennials? Am sure most of them are not millennials? And how exactly we identify specific user preferences based on their age / cultural preferences or geo political features. Even though there is enough evidence from data, path forward is to reduce dependency on business knowledge and bring data knowledge to forefront.
Key Watch out Item: Year 2020 will have lot of frameworks and slides replacing iProducts to aProducts (signifying the shift to Augmented Analytics).
Another key AI lever is Augment Data management as called out in Gartner trends, however for benefit of planning, it would make lot easier for Organizations to combine Augment Data Management with Augmented Analytics, as one cannot be accomplished without other.
2) Natural Language Processing and Unstructured data processing: It’s no secret that NLP is one of the gold mine or (new Oil) which hasn’t been tapped fully by many organizations. This year it will be of more focus with significant advances available like BERT and NVIDIA Megatron. Also there are more tools now coming available for Auto NLP which will drive this adoption more efficiently.
3) Graph Analytics: Graph databases finally have crossed the toughest trough in hype cycle and are reaching the plateau of productivity shortly. Most organizations have done POCs on Graph databases / use cases and have shelved them temporarily, this year we will see more focus on Graph use cases and most ML use cases will extend beyond traditional ML features. Top 3 use cases where graph will make difference is Customer 360, Fraud detection, Cyber Security. Other use cases are log analytics, identity management and Knowledge graphs will become mainstream and confluence pages will be powered by Graphs
4) Data Fabric: This will be a key rally point for organizations, even though it is over lapping with Hybrid cloud, it is important to align a seam less data fabric approach to integrate data from streaming, batch and API data along with structured and unstructured data. There is lot of innovation on this from NetApp, ever since the term is coined by them. Also other major product worth exploring for data fabric is MAPr
5) Compute Fabric: Often over lapped with Infrastructure strategy, Compute Fabric is going to be a key area of focus in this year, GPUs are becoming main stream with more increased focus on FPGA accelerators. Most of all ML frameworks like PYTorch has Quantization as high agenda also NVIDIAs RAPIDS is making great strides in becoming main stream. FPGA (field-programmable gate array) Accelerators are also now entering productivity plateau bypassing hype cycle and Quantum computing is now reaching its Hype cycle. Compute Fabric is so critical for future as best ML algorithms of future will leverage combination of CPU, GPU and FPGAs seamlessly to complete one processing cycle. These Models can no longer be executed in Silos and Next gen ML Models need seam less compute fabric.
6) Explainable AI: Explainable AI (XAI) refers to methods and techniques in the application of artificial intelligence technology (AI) such that the results of the solution can be understood by human experts. It helps us to develop interpretable and inclusive machine learning models and deploy them with confidence. Deep learning Models are already main stream and there is an increased demand from Regulators to make the models more explainable. There is a significant focus on making the Deep learning models from Black Box to more explainable Models. Leading packages which are helping in XAI are SHAP, AIX360, What — IF Tool and many neural network packages are coming up to help unbox the Deep learning Models.
7) Hyper Automation with AI and RPA: Hyper automation deals with the application of advanced technologies, including artificial intelligence (AI) and machine learning (ML), to increasingly automate processes and augment humans. The RPA bots and humanoid bots needs more integrated AI knowledge through ML / Deep Learning or Edge Computing to cater to the business needs. Hyper automation extends across a range of tools that can be automated, but also refers to the sophistication of the automation (i.e., discover, analyze, design, automate, measure, monitor, reassess.)
8) Citizen Data Science a.k.a Democratization of AI: Citizen Data Science is one of the common buzz word for last few years which allows everyone from organizations to build models rather than restricting to Data scientists only. Similar to how Tableau transformed by bringing business insights to everyone, Data Robot, H20 Driverless AI and Datanami are transforming the development of AI Models. In order for these frameworks to meet the business need, they need to be augmented with in-house feature store to bring business insights to every desktop.
9) Cyber Security: AI in Cyber Security was most talked about over past few years, but the level of penetration it achieved in commercial organizations is far less than expected. Knowledge Graph, network monitoring and Identify management are some of the few use cases which are addressed leaving huge scope for further expansion
10) Conversational AI: Chat Bot are run away hit and accomplished the level of automation using combination of AI and ML, whereas voice Conversation bots are still not at the maturity level and need a significant improvement to compete with Human interactions. However it is not feasible to replace bots in current scenario and what is needed is significant augmentation with custom ML models to understand the customer preferences rather than relying completely on technology solutions alone.
P.S: I understand that this article is not exhaustive and I will write few more articles on each of these trend to explain the most recent innovations on these levers. Also as mentioned, the key parameter would be to define the maturity level of each of these levers which I will write in upcoming articles. Looking forward for your feedback and suggestions.
References:
https://www.gartner.com/smarterwithgartner/gartner-top-10-data-analytics-trends/
https://www.forbes.com/sites/cognitiveworld/2019/07/23/understanding-explainable-ai/#74cfedfa7c9e
https://venturebeat.com/2020/01/02/top-minds-in-machine-learning-predict-where-ai-is-going-in-2020/