Comparing the AI “Training Set” with Human Conditioning

I think that as much as the creators of AI try to distinguish their work from human nature, it is remarkable the extent to which the development of a language model mirrors the development of a person.

A person, after all, is an apparent identity that emerges in the mind and body of a human being after it has been exposed to all sorts of stimuli and information, some of it random, some of it deliberate, which we call conditioning.

With AI a key element to its success or failure is the “training set” – the data that is used to “to teach the model the relationship between the input data and the desired output, enabling it to make accurate predictions or classifications when presented with new, unseen data.”

This is certainly what parents and teaches do when they try to prepare a child for “real life.”  The hope there is also that when the child grows up and is presented with new data, it will be able to handle it and come up with the desired output.

In human terms, desired output can be seen as appropriate behavior and success in life – however that may be defined.  If the conditioning is heavily religious, it is easy to see that the desired behavior is obedience to certain norms.

Almost all of the conditioning for humans in today’s society involves seeing oneself as a separate entity, often in conflict with others and a hostile world – especially if those doing the training have been traumatized.

Basically, in both cases it’s “garbage in, garbage out,” with the training set a tiny microcosm of the actual world that the AI – or the human – is expected to navigate.  Moreover, if the AI is conditioned to favor certain types of people or behavior its output will be skewed in the same way a human learns prejudice.

For AI to remain safe, its training sets’ scope and content must be closely monitored. For a human to find fulfillment, his or her conditioning must be examined, seen through and often jettisoned.

It’s another clear area where the development of AI could hopefully serve to enable humans to more fully understand themselves – and eventually reconnect with the natural world out of which we all come.

What is a training set in AI and who determines its scope?

In the context of AI, a training set refers to a collection of data that is used to train a machine learning model. It consists of input data, often referred to as features or predictors, and corresponding output labels or target values. The training set is used to teach the model the relationship between the input data and the desired output, enabling it to make accurate predictions or classifications when presented with new, unseen data.

The determination of the training set’s scope depends on various factors and stakeholders involved in the AI development process. Here are some key points to consider:

1. Data availability: The training set’s scope is influenced by the availability of data that is relevant to the AI task at hand. The data can come from various sources such as public datasets, proprietary data, or data generated specifically for the purpose of training the model.
2. Task requirements: The scope of the training set should align with the requirements of the AI task. For example, if the AI model is intended for image classification, the training set should contain a diverse range of images representing the classes the model needs to recognize.
3. Domain experts: Domain experts or subject matter experts often play a crucial role in determining the scope of the training set. Their knowledge and expertise help identify the relevant features, labels, and potential biases that should be considered during the training process.
4. Data labeling: If the training set requires labeled data (supervised learning), the scope of the training set can be influenced by the labeling process. Annotation guidelines and the expertise of the annotators can impact the quality and comprehensiveness of the labeled dataset.
5. Ethical considerations: The scope of the training set should also take into account ethical considerations, such as fairness, inclusivity, and potential biases. It is important to ensure that the training set is representative of the real-world data and does not perpetuate or amplify existing biases or discrimination.

Ultimately, the determination of the training set’s scope is a collaborative effort involving data scientists, domain experts, stakeholders, and the development team, with the aim of creating a comprehensive and reliable dataset to train the AI model effectively.

So a training set is analogous to human conditioning – how one learns who one is and the rules from parents, school and society?

The concept of a training set in AI can be loosely compared to human conditioning and learning from parents, school, and society in certain aspects. Here’s a broad analogy to help draw connections:

1. Input data: In AI, the training set provides input data to the model, which serves as examples for the model to learn from. Similarly, in human conditioning, individuals receive input from their surroundings, such as parents, school, and society, which provides them with information and experiences.
2. Learning process: In AI, the model learns patterns and relationships within the training set to make predictions or decisions. Similarly, humans learn from their environment through a process of observation, imitation, instruction, and experience. This learning process shapes their understanding of themselves and the rules of society.
3. Generalization: AI models aim to generalize their learning from the training set to make accurate predictions on unseen data. Likewise, humans generalize their learning from their upbringing and education to navigate new situations and make informed decisions.
4. Bias and societal influence: AI models can inherit biases present in the training data, and efforts are made to address and mitigate such biases. Similarly, human conditioning can be influenced by biases, cultural norms, and societal expectations, which individuals may internalize to varying degrees.

However, it’s important to note that the analogy has limitations. AI training is focused on pattern recognition and optimization, while human conditioning involves complex cognitive, emotional, and social processes. Human learning is more dynamic, context-dependent, and influenced by individual agency, critical thinking, and personal experiences.

While there are conceptual similarities, it’s crucial to understand the distinctions between AI training and human learning, as they operate within different frameworks and contexts.