Implementation of AI in Digital Transformation and Business

Abstract

This is a short overview of AI's widely used services and my experience in implementing them as a part of Digital Transformation for various businesses. November, 2023

?Introduction into AI

Nowadays, we are witnessing the widespread use of Artificial Intelligence (AI) throughout industries. The trend has become so popular that we can’t imagine our life without the help of our irreplaceable advisers: ChartGPT, Alexa, Siri, Copilot, Bing AI, Google Bard, Jasper Chat, and Personal AI to name a few…

I have been working with Neural Networks since 1998, when under the guidance of my professor Dr. Vladimir Khandetskiy, I created my first Backpropagation pure C++ written Neural Networks which were taught to decipher drawbacks from the defect signals for composite materials used in the Space science. The scientific mathematical principles have not changed much since then, but the technological advances have made a big step forward…

AI Lexicon

Some people do not possess a clear understanding of the AI lexicon. First, let’s decipher some terms in a very simple definition.

AI is the ability of a Software Application or Service to exhibit the behavior naturally appropriate to a human being.

ML (Machine Learning) is a part of AI which can be conditionally divided into 3 major groups:

1.?????? Supervised learning algorithms

• Regression algorithms
• Classification algorithms

2.?????? Unsupervised learning algorithms

• Clusterization

3.?????? Reinforcement Learning algorithms.

Supervised learning uses labeled data to train a model to make predictions or classifications, while unsupervised learning uses unlabeled data to find patterns or similarities in the data. Reinforcement learning does not use any data but rather learns from its own experience and feedback (like self-driven cars).

Deep Learning is a part of AI that imitates the functioning of Neural Networks (similar to a human brain). It can use the Supervised, Unsupervised, Semi-supervised (can be useful when there is a large amount of unlabeled data available but labeling them is costly or difficult. By using both types of data, semi-supervised learning can improve the performance and accuracy of the model compared to using only labeled data), or Reinforcement learning.

Generative AI is a part of AI that can use Large Language Models (LLM) or Foundation Models (FM) to create new content: text, images, audio, video, code, or any other syntactic data.

LLM is used to generate new text in natural language and other Natural Language Processing (NLP) tasks while FM is a large model which is trained to spot out patterns in input data.

While generating content, Generative AI can use both predictive and prescriptive models. The difference between them is that the former uses historical data and statistical methods to forecast what might happen in the future, while the latter uses optimization and simulation techniques to suggest the best course of action to achieve a desired outcome.

For example, a predictive model might use past sales data to estimate how much demand there will be for a product in the next year, while a prescriptive model might use the same data, plus factors such as costs, historical demand vs supply, and goals, to recommend how to price the product, how much inventory to order and how to allocate resources…

Predictive models can help identify trends, patterns, etc., Prescriptive models can help optimize decisions and outcomes, but they require more data, computation, and expertise to implement.

Cognitive AI Services set of cloud-based APIs that you can use in AI applications and data flows. They provide pre-trained models that are ready to use in your applications, requiring no data and no model training on your part.

Typical Cognitive AI Services capabilities widely used in practical apps are:

·???????? Vision. It includes facial recognition and optical character recognition (OCR), recognition of objects and people, and classification of images.

·???????? Speech. Functions as speech-to-text and text-to-speech, action items, drawing conclusions, automating video captioning, and creating audio content.

·???????? Language. Not just translation but also sentiment analysis, key phrase extraction, and opinion mining. ChatGPT—is a good sample of the natural and intuitive AI interface with natural language capabilities.

·???????? Decisions. Includes anomaly detection, automatic content personalization, automated content moderation for text, image, and video applications, robotics, and self-driving cars.

?

How to implement AI as part of DT in Business

Let’s investigate what we typically need to implement AI in Business.

1.?????? Reveal the business needs and pain points that can be addressed by this technology.

2.?????? Define your company goals related to the implementation of AI. To do this you need to:

• Create a roadmap for AI implementation strategy.
• Uncover metrics, KPI, OKR.

3.?????? Assess your company’s AI readiness. You need to evaluate your current capabilities and resources, such as data, infrastructure, skills, and budget. Assess risks, come up with mitigation plans, gap analysis, and use expert opinions on implementation plans.

4.?????? Prepare a Business Case and Solution plan for the technical project implementation.

5.?????? Get top management buy-in on the Project Plan, related to budget, schedule, goals, and major milestones.

6.?????? Prepare Corporate Culture to embrace new technology. You also need to foster a culture of innovation and collaboration and encourage continuous learning and feedback among your employees and stakeholders.

7.?????? Integrate AI into selected tasks and processes within your organization. You need to choose the areas or functions where AI can add the most value and impact, such as customer service, marketing, sales, operations, or product development.

8.?????? Learn from your mistakes and aim for AI excellence. You need to monitor and evaluate the performance and results of your AI initiatives and adjust or make improvements as needed.

Practical samples of AI inculcation

Many companies start a long Digital Transformation with AI journey by instigating Digital Optimization (see more about it in my previous article “Digital Transformation vs Digital Optimization”).

For example, many Outsource software development companies’ managers spend about 50% of his/her working hours on meetings with clients of different projects. Surely, they need time to prepare meeting minutes after this (or let someone do it for them).

Meeting minutes allow not only for preserving facts that were agreed upon but for managing clients’ expectations, assigning responsible persons, and action plans, drawing vital conclusions, etc.

The cost of such operations for a company is enormous. Let’s say that for the whole day, it will take at least 1 hour to make it up. So, it will take approximately 12.5% of the working time of those managers.

In Microsoft Teams by implementing “Start transcribing” functions, AI can generate Meeting minutes automatically with appropriate action items, Responsible persons, and even conclusions…In a matter of a few seconds…

The cost of an annual subscription is not even compared with the cost of the time the manager had to spend on composing those letters before, not to mention a 12% potential boost in productivity. IMHO nowadays, it is essential for Outsource Software companies to have this tool. It is a sample of a Digital Optimization technique.

The implementation of AI full-fledged chatbots will surely enable 24x7 access to your services and products by clients, raise CSAT (customer satisfaction) and NPS (Net Promoter Score) levels, help resolve typical problems for clients and free resources for other work.

The more sophisticated and byzantine AI technique implementation approaches in business include the creation of AI-enabled business functional use of roadway segments in streets of Digital Twins of big cities (proprietary created and patented by Joe Conroy and his company STS) or using AI technology for Authentication based on DNA Information in FASTA coding and biometric information, or AI-enabled research of DNA of the person to see his/her propensity to some diseases or health problems…

Apriorit company has used AI models to recognize different types of skin cancer, detect follicles' anomalies for healthcare, detect unusual traffic for cybersecurity, recognize faces on videos, assess vehicle damages in insurance, and many more…

The implementation of Face Recognition Technology is planned in Hyderabad (India) on a pilot basis but implementing this technology nationwide has met with big concerns from human rights activists because it might violate the privacy of a person or his/her human rights.

Pitfalls and issues in AI implementation

During the implementation of AI one should take into consideration Responsive AI principles:

1.?????? Fairness. There should be no bias in data training. For example, if an AI system is trained to process CVs only from highly industrial regions it may refuse candidates from distant places because of this…

2.?????? Reliability, high morale policy principles of safety. The probability of the result of the important decision should be verified by humans. In the case of self-driving cars, the traffic rule might be violated to preserve human life as the most important value for AI. Safety is an important delimiter of how the system should work even in untested or untrained situations.

3.?????? Privacy and safety. No Data breaches are allowed because of exposing sensitive information by AI Services. GDPR, CCPA, HIPAA, GLBA, etc. are obligatory standards to be observed by AI.

4.?????? Transparency. The models and algorithms used should be transparent for employees, so everyone understands why AI proposed that option. It prevents misunderstanding or misusing of the outcome and enables valid estimations of AI capabilities.

5.?????? Accountability. The company that created the AI service is accountable for the results. For example, Microsoft claimed that the more you chat with Tay, the smarter it gets. Unfortunately, within just 24 hours of launch, people tricked Tay into tweeting all sorts of hateful, misogynistic, and racist remarks. That was the reason to shut down that service. AI should be empathic, and Tay violated the Inclusiveness principle as well.

6.?????? Inclusiveness. AI systems should be designed and developed with the input and feedback of diverse and representative stakeholders, and they should be accessible and usable by a wide range of people.

It should be noted that AI is not a 100 %-fit-all solution. Its usage should be accurately weighed to include all the pros and cons of the solutions. That is why the advice of experts can be very valuable, especially for big digital transformation initiatives.

Summary and conclusions

As a recent Forrester study summarized, “AI adoption is no longer an emerging trend, with 73% of data and analytics decision-makers building AI technologies and 74% seeing a positive impact in their organizations.”

Globally, businesses are projected to double their AI investments from 2021 to 2025.

In case you’re the owner of the business and would like to empower and strengthen it with Digital Transformation based on AI and other technologies, or have any questions please feel free to contact me directly. I would be happy to help you with your business needs and may lead this transformation for you.

Ihor Antoniuk, skype: igorantonyuk

Email: [email protected]

Tel. +48 783752406

#AI #ML #DigitalTransformation #Apriorit #STS