Collaboration between Humans and LLMS
The
concept of 'user' when referring to a human interacting with an AI-powered
chatbot or Large Language Models in general [29], or digital tool is limited.
This is because the relationship is established through typical activities that
define a relationship between people: the exchange of information, questions,
answers, doubts, reconsiderations and discoveries, etc. While one can use the
chatbot in a broad sense when carrying out operations directed at that
particular object, when dealing with a digital communication tool, things
change and the 'use' becomes 'cooperation' [27]. Cooperation means working
together on an issue that will change according to different points of view and
capabilities. Cooperation can be viewed as a process that occurs over time and
involves the interaction of multiple actors. Imagine education, for example,
where at least three main groups interact: students, teachers and parents. In a
cooperative process, what is important is the mutual exchange of experiences,
feelings and ideas. This exchange brings about radical change for everyone
involved. Highlighting the fact that human-digital machine interaction is a
'collaboration' rather than mere 'usage' opens up a more complex kind of
interrelation, where newness, unexpectedness and creativity intervene in the
entire process and bring about change in the actors involved. Similarly to how
intellectual/emotional growth in education is not simply a matter of storing
more information in the brain, it can also be assumed that there is a
reciprocal enhancement of the actors. Consider, for example, the function of
assisted writing provided by AI tools: frequent interaction will enable the
human user to improve their language skills, while the machine will learn how
to enhance poor or ordinary language. This increase in skills and knowledge is
not the result of a set of issues, but the consequence of a systemic interrelation
that does not belong exclusively to humans or machines. It is exactly the same
with education, where the learning process depends on everyone involved;
however, it cannot be viewed as the superimposition of different layers.
·
The role of the observer
Emergent
properties arise from frequent interactions between systems. One of the most
underestimated yet relevant factors in the case of an AI tool is the role of
the observer. It is generally accepted that action is the most effective method
of bringing about change. Indeed, in order to experience the refreshing
sensation of cool air on a hot summer's day, it is necessary to open a window
(while hoping that the external temperature is not excessively high) or to take
a walk in the park or by the sea. It is difficult to accept that merely
observing a window or photograph is sufficient to experience the anticipated
sensation of wind. Indeed, this is the case. It is evident that the door of our
house is not designed to open automatically, and it is reasonable to conclude
that it would take a considerable amount of time for this to happen.
Nevertheless, observation engenders change. Consider the process of reading a
book, which is an activity that necessitates a certain degree of illumination.
The effect of the light rays on the paper is gradual and incremental, resulting
in the paper gradually yellowing. In certain cases, an individual may be
observed by the observer, for whom that observer's opinion is of importance.
This may be a friend, or a child. When asked to provide their opinion, the
observer may disapprove of the individual's attire. It is conceivable that they
may opt to alter their appearance. It is evident that the aforementioned
factors necessitate the conclusion that the observer is not a neutral spectator;
rather, they act as an activator of processes [28-30]. However, it is erroneous
to consider this perspective as purely subjective, as argued by Fields and
Urbani Ulivi, who contend that the observer's constructivism should not be
reduced to relativism, that is, to the arbitrary nature of perspectives.
In
which sources may this evidence be found? Quantum physics has provided the
knowledge that has rendered the application of classical physics to the
interpretation of life phenomena insufficient and inaccurate. Without
digressing, let us proceed to the point of interest by briefly recalling a
renowned paradox, "Schrödinger's cat", named after the physicist and
father of quantum physics, Erwin Schrodinger [31]. The following hypothesis is
hereby proposed: if a cat were to be situated within a container in which a
decaying atom were to activate a lever that resulted in the release of cyanide,
a substance with the potential to cause the cat's demise, then this would
result in the cat's death. In summary, the events within the container are
distributed between the probabilities of the cat being alive and dead. This is
impossible in classical (deterministic) physics, but not in quantum
(probabilistic) physics. The cat, regarded as a macroscopic living being in
comparison to the atom, is in a state of quantum correlation with the atom,
thereby adopting its characteristics. Given the fifty-fifty probability of the
atom either decaying or not decaying, it can be deduced that the cat's
probability of being alive is also fifty per cent. The transition from
indeterminacy (alive and dead) to determinacy (alive or dead) occurs
immediately upon opening the box. The act of opening the box is tantamount to
observing. According to the aforementioned standpoint, reality is not created
by the observer; rather, it is the observer who is created by reality. The
ultimate authority rests with the observer. If quantum physics is concerned
with very small systems quanta, the elementary and indivisible quantities of a
given magnitude the field of study known as the life sciences deals with
macroscopic objects/subjects (for example, the cat, the car, etc.). Beyond the
differences in size, there remain two common, interrelated features: the
indeterminacy of the phenomena of life, of the objects of reality, and the role
of the observer in constituting them.
·
Neuronal similarity between human
subject and chatbots
Last
but not least among the processes identified as the origin of emergence in
human/chatbot interactions is the similarity of brain functions. The capacity
for awareness, which has been viewed as the basic level of consciousness in
machines and recognized by digital devices in quasi-Socratic dialogue, is
closely related to the way data is processed. This environment exhibits certain
notable features of the brain, which will be discussed in this paragraph. It is
important to emphasise that these similarities facilitate the quasi-human
connection between the two interacting parties. The central processing unit
(CPU) can be regarded as the 'brain' of the digital machine, with the capacity
to process binary data using a set of instructions [32, 33]. The Central
Processing Unit (CPU) carries out arithmetic and logical operations in order to
manipulate data stored in memory devices. Examples of such devices include
Random-Access Memory (RAM) for short-term use and Hard Disk Drives (HDDs) or
Solid-State Drives (SSDs) for long-term storage. The machine is designed to
convert the binary data back into a human-readable form for display on output
devices, including monitors, printers, or speakers. The degree to which
machines are capable of enhancing their understanding and cognitive abilities
is contingent upon the efficacy of the underlying algorithms that govern their
operations. The aforementioned algorithms process vast amounts of data in order
to identify patterns, and thereby make decisions. It is noteworthy that these
processes are analogous to the manner in which humans acquire knowledge through
experience. Three distinct processes can be identified: machine learning, deep
learning and reinforcement learning. Machine learning is a subset of artificial
intelligence in which computers are capable of acquiring knowledge from data
without the necessity of being explicitly programmed through the application of
algorithms and statistical models. Machine learning techniques have been
demonstrated to facilitate the construction of predictive models and the
implementation of data-driven decision-making processes. Deep learning can be
defined as a specific instance of machine learning that employs neural networks
for the purpose of data-driven learning. The utilisation of deep learning
algorithms has proven to be particularly efficacious in the analysis of
voluminous and intricate datasets, with these algorithms having been employed
in a plethora of applications, including image and speech recognition, natural
language processing, and machine translation.
The
employment of deep neural networks, comprising multiple layers, facilitates the
execution of sophisticated tasks, including image and speech recognition. In
the domain of artificial intelligence, reinforcement learning is defined as a
subcategory of machine learning that focuses on the development of algorithms
that enable agents to interact with their environment in a manner that
optimizes the value of a specific reward, or reward function. The employment of
machine learning in training computer programs to play games and control robots
is a notable application, and the potential for its application to a diverse
array of real-world problems is significant. Within the paradigm of
reinforcement learning, the learner's decision-making process is typified by a
cycle of trial and error, a principle that exhibits a notable parallel with the
human learning experience. The predominant trends in the domain of augmenting
the cognitive capacities of machines are centered on the development of
computational artificial neural networks (ANNs), which are modelled on
biological neuron models comprising multiple interconnected nodes, designated
as "neurons". At this point we are reaching the core of the paper:
the quasi-Socratic dialogue will be revisited in order to explore the discourse
and emphasise the margins of autonomous machine reflection.