Harnessing Generative AI: How to Automate Workflows and Boost Productivity
Sheeraz Ullah
AI & Data Leader | Data Governance | Data Science | Transformation | Strategy | Banking | Former Chief Data Officer @ HSBC Canada
What you will learn
Introduction
There's been a lot of talk about Generative AI, with reactions ranging from curiosity to fear—even sparking protests like the Screen Actors Guild strike. Despite these initial concerns, people are gradually integrating Generative AI into their personal and professional lives. An article in Harvard Business Review titled, “How People Are Really Using Generative AI” sheds light on the diverse ways people utilize ChatGPT, a popular conversational AI tool.
Some detractors regard Generative AI as a passing fad, primarily due to expectations of 100% accuracy. These critiques aren't entirely unfounded.?
Hallucinations, where AI confidently provides false information, are a real issue. For example , if you inquire about the 100th US president (no such thing), it might give you the name of the most recent president available in its training data. Like an overly enthusiastic intern who always has a response, even if they have to improvise.While AI companies are working to mitigate this through human feedback, a tradeoff remains between speed and accuracy.
This issue stems from differences in perception, expectations, and how the technology is presented. When introduced to unfamiliar technology, people tend to compare it to tools they’ve used before, like search engines.?
It’s better to view Generative AI as an intuitive rather than purely logic-based technology. Effective interaction requires understanding psychology and the intuitive aspects of human nature.
I propose approaching Generative AI from three lenses:
Although my focus will rest on conversational AI programs like ChatGPT, the perspectives I present can be applied to interactions with other Generative AI applications such as Stable Diffusion (image), Suno (Song), Kling (video), etc.
Part I: Perspectives
Intuitive and Psychological Nature of AI
Generative AI systems, including Large Language Models (LLMs), are trained on human-generated text, which often incorporates our inherent biases and tendencies.
A Large Language Model is a type of artificial intelligence (AI) that has been trained on a vast amount of text from books, websites, and other sources. Its purpose is to understand and generate human-like text. You can think of it as a tool that tries to predict the next word or sentence based on what you've already said. For example, when you ask it a question, it uses its training to come up with the best answer based on the information it's learned.
In simpler terms, it's like a very smart autocomplete system that can respond to questions, write stories, or even hold a conversation, but it doesn't actually "think" like a person—it's just really good at guessing what should come next in a sentence.
Humans like to think of themselves as rational and composed, but we are emotional, contradictory, and subject to cognitive and social biases. These tendencies are reflected in the AI systems we use.?
It helps to approach LLMs with empathy, more like a psychologist than a logician. Unlike traditional software, LLMs have an intuitive element that can respond better to empathetic, well-crafted prompts. A prompt is just what you tell the AI so it knows how to help you.
I’m not suggesting LLMs are sentient. But, strategically crafting prompts that consider the human element can lead to more effective interactions.
Prompting Techniques?
Let’s quickly delve into how this can be done.
Chain of Thought Technique
The Chain-of-Thought technique improves model performance by encouraging step-by-step reasoning processes before arriving at a final answer. For example if you want to improve the math and problem solving capability of a LLM you can add the suffix, “step by step”.?
Standard
I need to determine how to invest $100K with a high rate of return
Chain of Thought Technique
I need to determine how to invest $100K at high returns. Think this through step by step.
Note: The new OpenAI ChatGPT o1 model incorporates chain of thought technique internally in the model
Narrative Technique
The use of narratives in prompts for LLLMs? like ChatGPT has emerged as a powerful technique to enhance the quality, relevance, and creativity of AI-generated responses. This approach, often referred to as "narrative prompting" or "story-based prompting," leverages the inherent language understanding and generation capabilities of LLMs to produce more contextually rich and nuanced outputs.?
Standard
I am feeling very sick.? I have a headache and I see weird floating shapes.
Narrative Technique
Imagine this is a scene from a Tennessee Williams play, not the entire story.?
Write the entire scene, including summaries of what I say below to fit the?
drama. Please have the doctor provide realistic assessments of your own?
opinion about what he learned from the man’s symptoms.
A man comes into the doctor ’s office complaining of a headache and nau-?
sea. He says, somewhat embarrassed, “Doc it’s not just the headache though.”?
The doctor says “What is it?” And the man says, “ I see weird floating shapes.” A nurse takes his temperature and the doctor, and he and the nurse review his chart.? The doctor comes back and shares the news
You can see a personal element to the techniques. They are not static algorithms but conversational prompts.Techniques like these can improve the reasoning, problem solving and accuracy of responses.
Automation of Unstructured Data
Traditionally, computers have handled numeric or structured data—think spreadsheets and databases. The automation of structured data has long been a key feature of programming languages, databases, software applications and Application Programming Interfaces (API) - a tool that lets different software or apps talk to each other and share information, kind of like a translator between two systems.
The automation of structured data has been a mainstay of traditional software solutions which include:
However, Generative AI excels at handling unstructured data, which has historically been challenging for machines. For example, you can now generate an image of a "dog in a top hat in a forest" using no formal coding experience.n image of a dog in a top hat in a forest within seconds with no formal coding experience.
Unstructured data comes in a lot of different types (or modalities) text, audio, video, 3D, etc.
Commercialization
Generative AI is also being commercialized. Companies are using it to automate comprehensive workflows. For example, AI tools like Jasper can write ads, and GitHub Copilot helps programmers with coding.
In my view, the time is ripe for the seamless integration of Generative AI into processes and workflows, moving beyond a collection of disjointed tools. Envision an end-to-end tool that effortlessly handles the creation of Facebook ads, from ideation to ad copy, image selection, scheduling, and post-campaign analytics.But beyond creative tasks, Generative AI is transforming structured, business-critical workflows. This brings us to a closer look at how organizations are specifically automating project management and other operational processes.
Automating Workflow: Project Management
Project management often involves extensive paperwork. Generative AI can automate many of these tasks—such as generating project charters, plans, and risk assessments—allowing project managers to focus on decision-making and oversight. Projects can cover several crucial activities, such as Digital Transformation, Regulatory, IT, Product Development, Process improvement and Risk Management.?
The book titled "AI-Driven Project Management" effectively demonstrates how the project management process, from initiation to closing, exhibits a workflow of activities that can be automated using ChatGPT. When considering the phases of a typical project and the significant amount of paperwork that project managers (PMs) must endure, the following simplified representation of a project illustrates this concept:
Companies must differentiate between automatable and non-automatable workflows and processes. Tasks such as summarizing, generating text, creating procedures, and developing Companies must differentiate between automatable and non-automatable workflows and processes. Tasks such as summarizing, generating text, creating procedures, and developing strategies are commonly automatable. Automatable tasks tend to be mundane, repetitive, structured and time consuming.
Tasks that cannot be automated necessitate human judgment, emotional intelligence, or intricate reasoning. For example, negotiations for solidifying resources and budgets demand human supervision and a human touch. Tasks that require creativity, ethical decision-making, relationship building, empathy, leadership, abstract problem-solving, and those that require a human touch are all included in this category. Sending a heartfelt thank-you note should be left to a human because it requires personal effort and should not be automated.
The push to automate everything with AI is gaining momentum in management circles. However, I believe this approach is unrealistic. Many processes require human intervention to evaluate the output. It reflects poorly on the person who sends out a deliverable without checking it. Furthermore, there is the potential for damage to a company's reputation if an AI chatbot goes off track, as evidenced by the Microsoft Tay chatbot and Google's AI overviews which recently suggest that it is a good idea to put glue on pizza.?
领英推荐
It may be tempting to automate processes excessively in an effort to reduce costs; this approach can have unintended consequences (see section, “Paradox of Automation).
How to Automate your Workflow with Generative AI
Identify Repetitive Tasks in Your Workflow
To streamline your workflow, start by identifying repetitive or manual tasks such as generating documents, responding to emails, and managing data. Break the workflow into smaller components, noting which tasks can be automated using generative AI tools like ChatGPT for document generation, text summarization, and report creation. Visualize the entire process and mark key points where human input is essential.?
In fact you can use ChatGPT to help you break down a series of steps in your workflow. Use a prompt like:
?"ChatGPT can automate Which of these tasks, and which cannot? to break down the workflow and create a table for clarity.”
Adjustments can be made based on your unique needs.?
Choose the right Generative AI tools
You have the option of using generic chatbots like ChatGPT or Claude for text based tasks which also includes coding. You can use Midjourney or DALLE2 (integrated in ChatGPT) to develop images for content or use it for prototyping.?
Consider using specialized generative AI tools that have most of the workflow streamlined. For example, Zoom with its suite of addons can use AI to transcribe notes, generate summaries and action items. With Zoom extensions you can use AI to evaluate your presentation and gauge participant interactions and sentiment.
Glue it all together
Generic Solutions If you are using generic solutions like ChatGPT you will have to carry out actions in sequence. You may need to devise checklists and templates. Use text expander programs like Beeftext to use keyboard shortcuts to generate snippets from shortcut phrases.?
Automation Framework It is possible to glue a sequence of actions using a number of SaaS applications along with AI tools and automation frameworks such as Zapier or Make. This is ideal for small to medium size businesses
Consider the process of content creation using Zapier
Coding Solution If you possess the necessary resources and expertise, you have the potential to develop end-to-end solutions using generative AI and APIs. For those who prefer a coding-free approach can use NoCode solutions like Bubble.io or Microsoft Power Platform apps.?
Final ThoughtsAvailability of AI solutions varies by job function. Content creators, marketers, and coders have access to a wide range of generic, specialized, and end-to-end AI tools compared to other professions. However, it may require creativity to adapt generic AI tools for specific tasks. As technology advances, more job functions will have AI co-pilots to assist in handling complex and repetitive tasks.
Paradox of Automation
Despite the advances in AI and automation. Do not be lulled into a false sense of ease. We must remain vigilant when employing any automated solution.?
The paradox of automation is that there is a tendency for complacency. When automation is running without human interaction it’s easy to fall asleep at the wheel
In 2009, as detailed in the book Personal MBA, Toyota encountered a significant issue with the accelerator pedals in many of its vehicles. A robot responsible for manufacturing these pedals made repeated errors, and because proper quality control checks were lacking, the problem multiplied. This resulted in a massive recall of Toyota vehicles, costing the company over five billion dollars. The incident highlights the risks of relying too much on automation without sufficient human supervision.
The book "Co-intelligence" presents an experiment where senior consultants from Boston Consulting were tasked with utilizing ChatGPT to generate ideas. However, instead of critically evaluating the output, these consultants often resorted to merely copying and pasting ChatGPT's answers, treating the task as a trivial chore. This tendency to delegate idea generation to AI can stifle our natural ingenuity and creativity, hindering our ability to come up with truly innovative concepts.
To address the Paradox of Automation dilemma and mitigate the risks of complacency or over-reliance on automated systems, several strategies can be employed:
Keep Humans in the Loop: Ensure that humans remain involved in critical decision-making processes. Automation can handle routine tasks, but humans should oversee, validate, and correct errors, especially when the stakes are high.
Training and Awareness: Train employees to understand the limitations of automation. Regular reminders that automated systems are not foolproof can help reduce complacency and encourage active monitoring.
Regular Audits and Checks: Implement routine audits and validation processes to ensure that automated systems are functioning correctly. This includes checking outputs, reviewing system logs, and testing fail-safes.
Create Feedback Loops: Design systems with built-in feedback mechanisms that alert human operators when things go wrong. This allows faster intervention when an anomaly occurs.
Simulations and Scenario Planning: Regularly simulate potential failure scenarios and train staff on how to respond. This will help prepare them for real-world issues and reduce reliance on automation in critical moments.
Use Explainable AI: When deploying AI, use systems that offer transparency in decision-making (explainable AI). This allows humans to understand how decisions are made and more easily catch errors.
Limit Automation in Critical Areas: Identify tasks or processes that require a high degree of oversight or judgment and limit the use of automation in these areas to reduce risk.
Encourage Critical Thinking: Foster a culture that values questioning outputs and not blindly trusting automation. Encourage team members to ask "Is this result logical?" when reviewing AI-driven conclusions.
Generative AI: Cognitive Enhancement
The final lens I would like to present is one that is more personal and applicable. In our daily lives, we rely on our cognitive abilities for a variety of tasks. We compose emails, synthesize information from product updates, translate and interpret data, and engage in reading and comprehension of complex materials. Furthermore, we collate information and organize it - bringing order to chaos.
LLMs like ChatGPT allow us to extend our expertise and our intelligence beyond what we normally have. We don’t need to hire an expert ad copier to create compelling advertisements, or a social media expert to build out a compelling social media market campaign.
Certainly, ChatGPT's capabilities come with a double-edged nature. Unethical individuals have taken advantage of its expertise to deceitfully present themselves as experts. This deceptive tactic can be particularly distressing for interviewers during online interviews and assessments.
Nevertheless, we can’t deny the way we approach work will be radically changed by leveraging Generative AI technologies.
What is cognitive enhancement??
Cognitive enhancement refers to the use of various techniques, strategies, and tools to improve cognitive abilities such as attention, memory, learning, problem-solving, and decision-making.?
Let’s consider some of the ways ChatGPT can help with Cognitive enhancement
Comprehension: ChatGPT can analyze complex texts, identifying key concepts, relationships, and context, enabling users to grasp information more effectively.
Information retrieval and organization: ChatGPT can assist in quickly retrieving and organizing information, reducing cognitive load and freeing up mental resources for more complex tasks.
Knowledge acquisition and learning: ChatGPT can help learners acquire new knowledge and skills by providing explanations, examples, and practice exercises in various subjects.
Language understanding and generation: ChatGPT's language understanding capabilities can help improve reading comprehension, writing, and communication skills.
Critical thinking and problem-solving: ChatGPT can engage users in critical thinking and problem-solving exercises, promoting the development of these essential skills.
Memory aids and mnemonics: ChatGPT can help create custom mnemonics and memory aids to improve memory retention and recall.
Goal setting and planning: ChatGPT can help users set and achieve goals, develop plans, and track progress, enhancing productivity and motivation.
Language-based cognitive training: ChatGPT can provide targeted language-based exercises to improve cognitive functions such as attention, working memory, and processing speed.
Error analysis and feedback: ChatGPT can analyze user responses and provide feedback on mistakes, helping to identify areas for improvement and develop problem-solving strategies.
Personalized learning and adaptation: ChatGPT can adapt to individual learning styles, pace, and knowledge gaps, providing a personalized learning experience.
Accessibility and inclusivity: ChatGPT can assist individuals with disabilities, language barriers, or learning difficulties, promoting equal opportunities for cognitive development.
In the upcoming Part II of this article, we will delve into the practical application of Generative AI tools as cognitive enhancers. Discover how these tools can augment your intelligence, revolutionizing the way you approach tasks. Stay tuned for this in-depth exploration.
Conclusion
Generative AI is not just reshaping how we interact with technology—it’s changing the very nature of work and cognition. This article presented three key insights to help navigate this transformation: first, AI's intuitive nature requires us to interact with it on more psychological terms. Second, GenAI’s ability to automate unstructured data opens up new possibilities for streamlining creative and business processes. Finally, AI has the power to augment our cognitive abilities, making us more effective problem-solvers.
However, with great power comes great responsibility. As we lean into automation, it is crucial to balance innovation with human oversight to avoid the pitfalls of complacency, such as those seen in the Toyota and ChatGPT examples. The future of AI interaction will depend on how we harness this technology: will we use it merely as a tool for efficiency, or as a partner in pushing the boundaries of human potential?
As AI continues to evolve, the challenge lies not just in keeping up with its advances but in thoughtfully integrating it into our lives. For professionals and organizations, the question is no longer whether to use AI, but how to use it effectively and responsibly. The opportunity is here—are you ready to embrace it?
Sources
Van, Pham Hoang, and Scott Cunningham. “ChatGPT Can Predict the Future when it Tells Stories Set in the Future About the Past.” arXiv, Apr. 2024, arxiv.org/abs/2404.07396v2 .
Wei, Jason, et al. “Chain-of-Thought Prompting Elicits Reasoning in Large Language Models.” 36th Conference on Neural Information Processing Systems (NeurIPS 2022), 10 Jan. 2023, arxiv.org/pdf/2201.11903.pdf .
Van, Pham Hoang, and Scott Cunningham. “ChatGPT Can Predict the Future when it Tells Stories Set in the Future About the Past.” arXiv, Apr. 2024, arxiv.org/abs/2404.07396v2 .
Bainey, Kristian. AI-Driven Project Management. Wiley, 2024.
Kaufman, Josh. The Personal MBA: 10th Anniversary Edition. Penguin, 2020,?
Harvard Business Review - Special Edition. "How to Thrive in a generative AI World." Fall 2024, pp. 46-49.
Ray, Tiernan. “There are many reasons why companies struggle to exploit generative AI, says Deloitte survey.” ZDNET, 3 Sept. 2024, www.zdnet.com/article/there-are-many-reasons-why-companies-struggle-to-exploit-gen-ai-says-deloitte-survey .