JourneyToAI

@mgunzelmann1 To learn more try https://www.coursera.org/professional-certificates...

We have experienced greater success in AI journey when we target low hanging fruits like Demand Forecasting

I'd say in order to progress along the AI ladder and productionize your potential AI solutions, think about where you can bring in automation, especially with data insights. Data-Ops (Data Governance) and Model Ops infused together creates end-to-end lifecycle possible

Early deployment of Model and Continuous Monitoring & Refinement of the same is key to unlock the value of AI. It is better not to spend months of effort to build the model and then find that it is not aligned to the reality.

Start looking at existing tools and technologies in AI space, inclusive of understanding your data pipeline as well. AI is as good as your data. https://www.ibm.com/blogs/journey-to-ai/2020/02/se...

@mgunzelmann1 Build your own sandbox to explore AI to see what is possible...https://www.ibm.com/cloud/free

Understand the end to end lifecycle and play with Cloud Pak Experiences. Explore this flexible multicloud data platform with free access to a hosted environment. https://www.ibm.com/cloud/paks/experiences/cloud-p...

Along with the data quality, ability to explore/ self service the models on a data lake or virtualized data layer for the citizen data scientists in the platform leads to early adoption right from the business users

There is no one right answer but think about forecasting, optimization, prediction problems etc. and see if you can inject any intelligence and automation to it.

According to #IBM Institute of Business Value Study 60% of the companies see regulatory constraints as a barrier to implementing #AI..

@mgunzelmann1 Trust but verify, a new idea for AI

Explainability leads to trust on AI models

@mgunzelmann1 It has to be a transparent process and your platform must include this

Enhancing Fairness and trust in AI models has been crucial for Operationalizing AI models and putting them into production. For imperatives involving AI adoption would typically include Drift, Fairness, Explainability and Resiliency.

Four trust imperatives hindering AI adoption
Drift: How do we ensure data quality throughout AI lifecycle?
Fairness: How do we reject bias?
Explainability: How can we explain decisioning of how AI came to its conclusion?
Resiliency: How can we shield AI against cyber threats.

Metadata and Lineage often are two factors in explainability of models

@mgunzelmann1 If users don't trust AI, it will never be used. Amke use you can explain your decisions.

Model's transparency can be achieved using IBM OpenScale (https://www.ibm.com/cloud/watson-openscale)
and AIX360 (https://aix360.mybluemix.net/) across the 4 key dimensions of Model's interpretability - Local Explanation, Global Explanation, Counter Factual and Similarity

(edited)

When we talk about Drift, we need to ensure data quality throughout AI lifecycle, which involves inspecting and continuously drift in data consistency as well.

@mgunzelmann1 Join the community; https://community.ibm.com/community/user/cloudpakf...

Similarly Fairness/Bias of Models can be easily figured out using AIF360 (https://aif360.mybluemix.net/) and IBM OpenScale (https://www.ibm.com/cloud/watson-openscale)

In terms of Explainability, we have local and global explainability. There are algorithms like LIME, SHAP etc. and providing Contrastive explanations to aid in providing reasonable insights behind ML Models. https://www.ibm.com/blogs/research/2018/05/contras...

(edited)

A study done by BCG shows that only around 20% of #AI models get into production.

The cost of quality is the first for sure - as without cleaning the data the AI is handicapped

Trust on AI Models is key to success of AI Operationalization.

@mgunzelmann1 Its really only mathematics giving the computer to "analyze" data the way humans respond.

I agree. In order to turn AI aspirations into outcomes, organizations need to overcome three major AI challenges: data complexity, skills, and trust.

@mgunzelmann1 Now the computer can read and answer requests, more than simple calculations

According to MIT Sloan, data collection & preparation is the most time consuming and difficult part of AI.

Along comes speed of decision as well, because by the time the AI is implemented the changing business scenario can leave it unusable, thus the need of real-time AI

@mgunzelmann1 New techniques provide automation tools to reduce the time consuming processes.

It is necessary to reach a stage where we can trust our AI models and provide explainable insights. https://www.ibm.com/blogs/research/2019/08/ai-expl...

@mgunzelmann1 Keep in mind, we need to able to assure that AI is making"fair" decissions

@mgunzelmann1 AI is much more than models today.

Before deploying a Model in production one has to ensure all necessary due diligence - data quality, independent validation of model, model's interpretability and overall governance in Model Development Life Cycle

In order to transition from black-box to glass-box, it becomes quintessential to provide reasoning behind model outcome. Data Complexity, Skills and trust play a major role in this aspect.

@mgunzelmann1 https://www.crowdchat.net/s/364a7

In order to accelerate the AI Adoption, it is important to streamline and understand the AI ladder- Collect, Organize, Analyze and Infuse. https://www.ibm.com/analytics/journey-to-ai

@mgunzelmann1 In order to get AI right, you need a platform to address the entire process, Collecting trusted data, organizing and analyzing it, putting it into production and most importantly monitoring it over time!

@mgunzelmann1 It is a continuous feedback loop of improvement.

Also, from an educational standpoint, there are active courses on Operationalizing AI and deploying AI framework into enterprise. https://www.coursera.org/learn/ibm-ai-ladder-frame...

From IBM's side we propose 5Cs (Continuous Training, Continuous Validation, Continuous Integration, Continuous Development and Continuous Monitoring) for success of AI Operationalization - https://github.com/ibm-cloud-architecture/refarch-...

@mgunzelmann1 A great solution that allows people deploy only what they need and grow over time.

A platform that can support the entire process of Ingest-Store-Manage-Analyze-Visualize (ISMAV) can handhold in the journey through AI maturity

“Enterprise AI” is about solving sophisticated business problems in highly dynamic environments. This requires an understanding of well-defined use cases. IBM's Decision Optimization on CloudPak for Data provides prescriptive analytics - https://www.ibm.com/analytics/decision-optimizatio...

Yes absolutely. Every customers we are talking today in various industries want to exploit power of AI.

It is surely desirable to inculcate AI for sustainable revenue growth. Around 80% of firms expect the number of AI use cases to increase within the next two years.

According to #IBM Institute of Business Value Study of 13,000 global C-suite 80% of companies are planning large investments in #AI.

@gunzelmann1 AI is a great way to make your orgainzation Smarter, Better and Faster

@mgunzelmann1 The world has finally started down the path of AI

@mgunzelmann1 AI is really about allowing a computer to mimic human responnse rather than just zeros and ones