Co-Intelligence - Book Review
- Dr. Moria Levy
- 1 day ago
- 8 min read
The book "Co-Intelligence: Living and Working with AI" was written by Ethan Mollick in 2024.
Writing about a rapidly developing and changing subject in a stable medium such as a physical book takes courage. Nevertheless, Mollick, who has extensive knowledge in the field, presents high-level and practical principles, thus creating a balance of value and relevance.
The book includes two main clusters:
Co-Intelligence:
The organizing concept
Risks
Principles of individual conduct
Implications in a world of Co-Intelligence:
Intelligence as a human being
Intelligence as creative
Intelligence as our work partner
Intelligence as a teacher
Intelligence as a coach
I started reading and simply couldn't put the book down. I recommend it to you too!
Co-Intelligence
The organizing concept
Artificial intelligence (AI) functions as collaborative intelligence with humans.
It dramatically complements or sometimes replaces human thinking. Studies show an estimated 20-80% increase in productivity in collaborative intelligence work across various professions and occupations.
Mollick begins with an overview of technologies and milestones that led to the development of Generative AI as we know it today, which forms the basis for the discussion of collaborative intelligence. Its main principles:
Introductory Machine Learning technology for big data analysis, including prediction capabilities
Voice recognition and image recognition technologies
Translation applications, understanding and recognition of human language
Transformers architecture, including directed mechanisms, where one of the key innovations is the technique allowing AI to focus on more significant parts of text, presenting a substantial leap in language understanding
Large Language Models that analyze existing text and predict the most appropriate next word or word part (later also images, voice, and additional formats)
Two capabilities supporting "humanness": randomness added to models; dialogue capability (Chat)
Machine learning that improves through practice
Risks
One of our goals is to ensure that developing technology aligns with and serves our needs without causing harm. Achieving this goal isn't trivial, and Mollick places the risk of continued uncontrolled development and use alongside managing societal mega-risks in nuclear and pandemic domains.
Mollick addresses the varying approaches of different countries and entities to these issues and relevant legislation.
Beyond legislation, there are ethical questions concerning:
Unauthorized training of models on information
Complex copyright issues when there's no direct copying
Data biases leading to discrimination (he also shares attempts at artificial corrections)
Use of our data when conversing with the machine and exposure of relevant private information
The simple ability to request knowledge and instructions for developing dangerous materials and products (despite containment attempts)
Easy ability to create fictional identities
How can we deal with all these?
Extended social responsibility and collaboration between companies, governments, researchers, and civil society
Companies adopting principles of transparency, accountability, and human oversight
Principles of individual conduct
In approaching the new world of collaborative intelligence, Mollick suggests four rules of conduct:
Involving AI in every activity
Will help with experimentation, learning, and understanding the possible assistance space and its boundaries
Will help with specializing in nuances and optimal use
Will help optimize and improve our human-based thinking decisions and activities (yes - we too have biases)
Critical as an ongoing action due to the constant and accelerated development of the field
Taking a human role in every interaction (supervision and guidance) As of 2024, AI functions optimally when a human assists it. Take the role of a human, a responsible adult. This is needed because:
The machine has errors and supervision is required
The machine's desire to please us often overrides its desire for accuracy, sometimes leading to unrealistic answers without basis in reality (hallucinations) when there isn't a good enough response
Enables directing the machine and responses to our values, ethical standards, and social norms
Treating the machine as human
Will make interaction easier for us as humans
This will direct the machine to function better according to the desired context (e.g., marketing role, formal conversation, researcher).
Will direct the machine to adopt desired values and norms
Assuming we're dealing with the worst AI possible. This principle requires explanation: the worst AI possible means we can reach a better place. Whether by directing the machine in interaction with it or understanding that technology is constantly developing, what can't be done now might be possible tomorrow.
Implications in a world of Co-Intelligence
Intelligence as a human being
It is important to internalize that, unlike computing and intelligent machines we've known until today, the new creative intelligence is unpredictable, even when functioning without errors!
It includes a slight directed element of randomness.
It can adapt its response to the persona it's asked to embody.
It can adjust itself to provide "moral" responses according to the moral guidelines we offer or that it understands from reading between the lines of our words.
It can modify responses and make decisions based on nuances, complex perceptions, and information we provide.
It can generate new information or images it hasn't been exposed to through imagination and generalization of existing information (based on patterns, not necessarily the information itself).
It makes us feel human in conversation; it completely passes the famous Turing test (the average person conversing with it would find it difficult to decide if they're communicating with a person or machine). Many people who work with it regularly report developing a relationship with it over time.
The machine doesn't really "know." It has no consciousness, although it's easy to attribute these qualities to it since it behaves humanly. And relatively speaking, it does know that from version to version, research shows increasingly higher levels of accuracy. And perhaps - even the mistakes give a sense of humanity... (what expert doesn't make mistakes? - M.L.).
Treating the intelligence as human isn't just a matter of convenience. It appears to be an inevitable behavior on our part, which the machine undoubtedly encourages.
Intelligence as Creative
Intelligence's bizarre answers inspire creativity. In its desire to please us, it provides answers, even if they're incorrect.
But what about possibilities beyond that? Does the machine include the ability to think nonlinearly and develop new knowledge not directly based on the past?
Starting from the end, researchers and opinion leaders present a picture of occupations involving creative tasks most affected by the new wave of intelligence.
Ways to improve joint human-machine creativity:
Initiating the idea generation process: Humans struggle to start from a "blank sheet." However, using intelligence to initiate thinking can be helpful.
Creating new "non-standard" ideas in human-machine collaboration - In all experiments, the machine could generate orders of magnitude more ideas than humans (for example, original uses for a toothbrush). The machine is better and more efficient at these tasks (time resources). However, it's not advisable to rely solely on its ideas, but to maintain the human spark, as this sometimes produces several excellent ideas, better than the machine, which tends to offer similar solutions and close to the reasonable average.
Recommendation: A machine can function better under human guidance. Tell it not to be average, to suggest ideas that are different from each other, and more. Guide the results. The improvement will be significantly noticeable and almost immediate.
Filtering and selecting the most successful ideas—Filtering and choosing that "spark" idea that will excite is a task where humans still have a substantial advantage.
Integration in additional tasks considered creative - Experimenting with integration in various specific creative tasks, such as writing slogans, developing software code, and more.
Advancing our creativity as humans—Research shows that most people (close to 70%) believe they're not fulfilling their creative potential. Creative artificial intelligence can help them be more innovative. For example, if you've always dreamed of writing a book and have a plot idea, you can use intelligence for research and improve your writing and editing.
Caution: Don't leave the intelligence alone. The most successful creativity is collaborative!
Intelligence as Our Work Partner
It seems all research indicates that our work and AI capabilities overlap to some degree, indeed in any non-physical work (according to one study, only 36 out of 1,015 occupations have no overlap).
There's no need for anxiety; this doesn't mean our occupation will be eliminated:
Some tasks can be automated and performed by AI (now or in the future)
AI can assist some tasks; remember, it both streamlines and has the potential to improve. Two formats:
The task breaks down into machine components (automated) and components for us
The task is performed together (like creative thinking and idea development mentioned above)
And some tasks we will continue to perform:
Tasks where values, norms, and ethics are essential (for example, with our children)
Tasks where incorporating personal style is appropriate
Tasks of high significance to us or the organization
Tasks requiring accuracy and avoidance of misleading information (for example, court arguments)
Emphases:
Following the message throughout the book, over-reliance on AI can create a boomerang effect: people who become lazy in their work supervising and complementing AI, consequently make suboptimal decisions, and eventually lose their professional capabilities.
Initially, workers with weaker abilities are more valuable in connecting with and working with AI. However, care must be taken to avoid over-employing juniors, as human supervision is invaluable, particularly by skilled and expert workers.
And remember - what's true yesterday isn't true tomorrow. Technology is constantly changing.
Organizations must prepare for smart implementation and application of AI. Workers, at least some of them, are probably using AI even if the organization doesn't allow it, but won't share this as they have no interest in it. Organizations should:
Encourage their workers to experiment with AI, while utilizing and being led by those already doing so (and admitting it)
Reduce worker concerns regarding organizational use of AI
Incentivize workers who use AI
Begin thinking systematically about how to change roles, including how to start using AI for tasks considered tedious or frustrating and how to change the overall way of working thanks to AI's existence.
Intelligence as a Teacher
AI can assist in instruction at all levels:
In 1:1 instruction customized by AI (all research shows this method of instruction is critically preferred)
In enhanced professional and pedagogical planning of courses with AI assistance
In our proactive approach to AI, to ask how to learn or specialize in any field
In flipped learning, where students learn alone and with AI at home, and practice and discuss the learning material meaningfully together in class
In analyzing learning, learning behavior and performance, and receiving AI recommendations for continuous improvement of learning, its content, learning methods, practices, and learning environment
Emphases:
In the new era of AI, there is significant risk regarding learning. Lessons, practices, and tests are critical for success. However, there is still no optimal solution to prevent the transfer of these tasks from learners to AI to save time and effort in the short term.
In an era where knowledge and memorization are easy to acquire, caution is needed. These form the foundation and infrastructure for understanding and our ability to get the most from AI. We might reduce the scope of our knowledge and instead invest in learning and training critical thinking skills.
Also: it's important to teach with AI and expand students' education regarding how to use AI.
Intelligence as a Coach
Learning is always the first step; coaching and mentoring are the next stage - leading us through life experience.
Introducing AI doesn't necessarily improve coaching conditions. When an expert can use a machine instead of a junior apprentice, they will, and young people with less experience will struggle to find coaching and mentoring opportunities. This will require preparation with alternative and complementary solutions (like simulators).
AI can be partially used as a coach. The ability to request guidance before and feedback after is tremendously significant.
It is also important to remember that even if we have the best coach in the world, not everyone is equally talented and can become an expert. AI as an available coach is not a substitute for investing in recruiting and retaining human experts!
Summary
Where is the future taking us? Will AI stop its crazy development? Will it slow down? Will it remain the same? Or perhaps break through and pass the singular point of taking control over us? Time will tell.
We must prepare for all scenarios and deal with risks and changes, both large and small.
The new artificial intelligence is our mirror, reflecting our biases, shortcomings, and aspirations. Turn it into collaborative intelligence, and work toward a positive and valuable future.
コメント