<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Alexis Jacquemin on Medium]]></title>
        <description><![CDATA[Stories by Alexis Jacquemin on Medium]]></description>
        <link>https://medium.com/@alexjacquemin?source=rss-d8733079bda8------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*e4g7KjIioHTDjs71byGTFg.png</url>
            <title>Stories by Alexis Jacquemin on Medium</title>
            <link>https://medium.com/@alexjacquemin?source=rss-d8733079bda8------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 22:15:16 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@alexjacquemin/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[What is Risen Learning?]]></title>
            <link>https://medium.com/@alexjacquemin/what-is-risen-learning-b8634231469e?source=rss-d8733079bda8------2</link>
            <guid isPermaLink="false">https://medium.com/p/b8634231469e</guid>
            <dc:creator><![CDATA[Alexis Jacquemin]]></dc:creator>
            <pubDate>Fri, 04 Oct 2024 22:30:12 GMT</pubDate>
            <atom:updated>2024-12-19T01:02:46.863Z</atom:updated>
            <content:encoded><![CDATA[<p><em>On Creation</em></p><p>This article is both an attempt at popularization, on the subject of what Artificial Intelligence is; it incarnates also an exercise in innovation, for a novel form of Artificial Intelligence.</p><p>***</p><p>The history of Artificial Intelligence is recent, compared to other arts and techniques; except that the idea of Artificial Intelligence is not new. If we talk about it so much these days, and if we are at present achieving such impressive results, it is less due to the discovery of brand-new algorithms than to the computational prowess of machines…</p><p>Indeed, some algorithms, which are now run very powerfully, have their origins in the intuitions of Alan Turing in the 1950s. Moreover, the statistical and mathematical laws on which Artificial Intelligence is still based go back many centuries of research!</p><p>As a consequence, to understand how Artificial Intelligence works, it would be more informative to explain the early concepts of the 19th century than to describe the latest architectures of the 21st century. In this sense, we have drawn up a list of terms employed for this article’s plan. While detailing their role in contemporary Artificial Intelligence, we will try to highlight their limits, and we will outline some alternatives, those followed by Risen Learning.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/933/1*ytYjSaRiuInE7PNZzcTA8g.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/936/1*4ghSiBdqbYeuNvBVOar9Bw.png" /></figure><p>There are of course several species of Artificial Intelligence (AI) that have been used more or less in the last decades. Overall, AI remains a gender of Advanced Functions: you set information as input, and you get renewed information as output. What justifies the title of Intelligence is that unlike ordinary computer functions, we do not know a priori what we will eventually obtain; in the meantime, Intelligence has applied its own processing which, for various reasons, was not anticipated, or which sometimes keeps being incomprehensible afterwards.</p><ul><li>The first species of AIs called <strong>Symbolic</strong>, which until recently were the most widely leveraged in industry, operate with Rules. You define the rules at the start of the computation, and on account of a fair margin of uncertainty, a combination results that can be enlightening.</li><li>Secondly, there are the so-called <strong>Connectionist </strong>AIs — these are the ones that are being discussed a lot nowadays. As their name suggests, they are able to develop connections between the information consumed, which they analyze by assigning Weights to them, classifying them by relative importance; and this allows for a particular and meaningful hierarchy.</li><li>Under this species, we often place <strong>Machine-Learning</strong> that mobilizes more complex Data. In reality, one could argue that most Symbolic and Connectionist AIs learn from Data; yet their learning mode draws quite closely upon what is given to them as input, having to tell them how to learn. Although, learning AIs are likely to generate new variations of Data more spontaneously, this in line with two methods:</li><li>Sequentially, by iterating through actions — for example, to explore an environment, and whose trial and error improve their perception of obstacles or other opportunities. To do this, algorithms reinforce the consequences of decisive actions and make disappear those that are not, so their title of <strong>Reinforcement </strong>learning.</li><li>Gradually, by spreading the initial Data, in combinations in the form of layers, revealing interstice secrets, which opens up a vision that is both comprehensive and precise. The more these AIs produce multiple connections, the more their learning is said to be <strong>Deep </strong>since they perceive the information with more relief, hence their relevance.</li></ul><p>In all the cases cited, we could already establish a key principle of Intelligence, that of validating or invalidating part of the information that AI has at first glance, with the contribution of increasing or decreasing the value of the information that comes up; and if there is genuinely Data generated, it is only by virtue of a quantitative transformation of the primary inputs. Today, only four formulas are applicable:</p><ul><li>More = More,</li><li>More = Less,</li><li>Less = More,</li><li>Less = Less.</li></ul><p>Notwithstanding, from Data, it is presently possible to adopt another, more qualitative, approach to information, more differentiating, of the types:</p><ul><li>Interior = Interior,</li><li>Interior = Exterior,</li><li>Exterior = Interior,</li><li>Exterior = Exterior.</li></ul><p>Reflecting only the Plus or the Minus, AI encompasses pivotally decomposing the information, then recomposing a new arrangement. Preferably, as we will demonstrate below, bringing together the Data, under an Interior or an Exterior, amounts to prolonging the Presence of each qualified Layout, as well as emulating the Convergence of all the differences; without reducing them to the cutting by Sequences, or to the leveling of Gradients, of arbitrary sizes. For this reason, we name this form of AI of Emergence by creative rapprochement of Data: Risen Learning.</p><h3>I. Sequential</h3><h4>A. Cooccurrence</h4><p>Let us take an example right away:</p><p>1. a) The cat chases the mouse.</p><p>1. b) The cat catches the mouse.</p><p>In these sentences, delimited by simple punctuation — constituting a sort of compartment, there is a cooccurrence of “cat” and “mouse”, that is to say that the two words occur together in the same places, here these two sentences. What an AI targets to calculate mainly resides in the frequency of cooccurrences: how many times do words, objects, actions happen at the same moment?</p><p>Then, if we ask the AI:</p><p>2. What is the cat searching?</p><p>Having defined more generally that:</p><p>3. a) Chasing is a form of searching.</p><p>3. b) Catching is the end of a form of searching.</p><p>AI will be able to induce, knowing that 1.a), 1.b), 3.a) and 3.b):</p><p>4. The cat is searching for the mouse.</p><p>So, starting from a few cooccurrences of words in several sentences, AI would be competent to answer this question, proposing a meaning that humans understand.</p><p>Things get complicated when words end up together for extremely specific reasons:</p><p>5. a) The cat and the lion are felines.</p><p>5. b) The mouse and the antelope are mammals.</p><p>If we add:</p><p>6. The lion hunts the antelope.</p><p>And that we request:</p><p>7. What does the cat hunt?</p><p>Due to these cooccurrences, AI could potentially assert that:</p><p>8. The cat (=feline) hunts the antelope (=mammal).</p><p>Because there are cooccurrences of “cat” and “lion” on the one hand — as felines, and of “mouse” and “antelope” on the other hand — as mammals, the AI might mix up the hunting objects of the lion and the cat, so that it will write rather incoherent sentences!</p><h4>B. Separability</h4><p>This is why the second essential tool of AI is context separation: it tries to distinguish situations, in order to be more precise. First, we could situate the “cat” in France and the “lion” in Zambia; and if we stipulate:</p><p>9. The cat walks the streets of Paris.</p><p>10. The lion is kept inside a park in Liuwa.</p><p>…it is unlikely that AI tells us that the cat is hunting an antelope in Paris. It is enough to finely determine the context ahead of separating what makes sense for the cat or for the lion respectively; from then on, an AI will apprehend the different situations better.</p><p>Furthermore, a second separation could be applied regarding the qualities specific to cats and lions; for example, one could measure the weight of the cat and deduce that it would be incapable of attacking an animal that was several times its weight, such as an antelope!</p><p>That pushes us to utilize the criterion of weight to distinguish all the animals that the cat and the lion hunt; and beyond a certain threshold, we gauge whether or not the cat is able to hunt such and such an animal, which eliminates sentence (8) — for an antelope, with its imposing size, could not be the prey of a small cat.</p><p>Whether one decides on the basis of a place or a property, it comes down to making a distinction of correspondence to the data. There is a series of data on the cat and the lion, and from the knowledge of the adequate data, one grasps what belongs to the “cat” or to the “lion”.</p><p>However, it is conceivable that in the same places and for given species, a cat and a lion come to hunt the same animal, a large hare for instance. The data “hare” would then be common to the cat and the lion; consequently, a wild cat just as a lion would pertinently hunt a hare.</p><p>By multiplying the ambiguous data, mobilizing all the feline characteristics of the cat and the lion, a confusion would loom and menaces weakening the level of Intelligence, which leads us to now resort to another stratagem…</p><h4>C. Variance</h4><p>Some of the data about the cat or the lion is precise enough to help us distinguish them; the rest of the data is not precise across all contexts. A wild cat would live in the savannah, a lion might condescend to hunt a tiny animal; which assimilates one to the other.</p><p>In which case, one would either have to increase the data to the point of delving deeper into the context — “Does the lion have access to enough large prey?”; or one would gain by removing the data that derives into indecision instead — no longer considering the existence of the “hare”, in this case.</p><p>Albeit it would also be important to keep a set of data shared by all felines, and thus accurately differentiate the hunting techniques of the lion from those of the wolf, or of the cat from those of the dog, to take another example. Having a wide variety of data is not always a sad thing!</p><p>Depending on the problem that is posed, it is sometimes critical to draw the lion closer to the cat, within the feline family, compared to canids in particular. The methods of stealth, the speeds of movement, the organizations in packs, so many subtle variations that we value to recognize a cat, a lion, from a dog, a wolf…</p><p>Intelligence subsequently does not consist solely in counting cooccurrences, nor in separating data; it is finally appropriate to act on the variance of cooccurrence data, privileging them or relativizing them in light of the issue. In the table that follows, we have to intervene on several categories of data:</p><p>The Columns are the characteristics common to all felines:</p><ul><li>Places of life,</li><li>Weight,</li><li>Hunting techniques,</li><li>Etc.</li></ul><p>The Lines, or all possible samples of felines:</p><ul><li>Paris Cat,</li><li>Liuwa Lion,</li><li>Tanzanian Serval,</li><li>Serengeti Lion,</li><li>Etc.</li></ul><p>Few actions are possible when we ask ourselves a question intelligently: we can add or remove a column, consider felines only through their living spaces, without being exposed to confusing their prey according to their weight; alternatively, we could try to give more importance to one or the other column, by weighting them: if the living space is 1000 times more decisive than the hunting technique, a Parisian cat will certainly not hunt a hare.</p><p>We can also add or remove lines, keeping only the felines that cover together or not the same territories. Or, choose to attribute more or less weight to this or that line, estimating that the AI reasons in particular on cats of an average size, or that it extends to future corpulences…</p><p>These are the main operations that are commonly applied to data, thereby enhancing an AI; and in a sense, these are as many Sequences that are cut out from the data to make them intelligent. Which ultimately signifies that Intelligence is deployed above all in the mode of implementation of Sequentiality, and which we call Graduation.</p><h3>II. Gradual</h3><h4>A. Probabilities</h4><p>So far, we have thought more about the grammatical and logical intelligence of statements; and this constitutes, in fact, a major part of the performances that Connectionist AIs oppose to Symbolic AIs. Competing with a treatise of grammatical rules carefully taught, after mere assimilation of tons of books, Connectionist AIs triumphantly win; by the unique distinctive enumeration — ergo intelligent, of cooccurrences, this on condition that these readings are sufficiently varied…</p><p>And that is where the problem lies: Connectionist AIs are only interested in things that are repeated many times, though an idea that only appears once will probably not be taken into account! Anything that is singular is effectively improbable, and improbability implies that it does not make sense to everyone, and it will not be retained by the AI. Connectionist AIs only succeed in generating connections if the variance of ideas is probable, provided that it is widespread.</p><p>The probability, in this way, ensures that the machine is recognized as intelligent in front of ordinary mortals: understanding what an AI says is proof that it interprets the same meaning as us, to the strict extent that it states what a plethora of human beings have already formulated before. In speaking, an AI, in a manner or another, boilerplates sentences, or pieces of sentences, that it has encountered somewhere, many times; otherwise, in trying something new, it would be considered delusional for being irregular.</p><p>Finally, for Connectionist AIs included, their Intelligence is reduced to respecting rules, in spite of the fact they were learned by observing their recurrence in a broad range of contexts. It is as such accepted that there is always a more or less beneficial play with the rules for a modest expectation, namely, to align with the average usage of these. The presupposition remains that any intelligence of situations is based on rules that we firstly see being necessarily applied by humans, before claiming to be probably recovered by machines.</p><h4>B. Central Limit Theorem</h4><p>Even when a machine learns from its environment, it exploits sensors that humans have previously tuned, with the goal of matching the limit of probable situations. In statistics, for a long time now, this central limit theorem has been worn out: in the middle, the most favorable context, that is, the one that forms a balance between too much and not enough — like a car driving on ice: going neither too fast nor too little; and around this center, all the possibilities that are all the less relevant the further they move away from the theoretical ideal.</p><p>The underlying assumption is that the perfect value is located in the middle of an equilibrium, and that all around are continuously spread out the degraded alternative values; the most distant peripheries being the falsest, while we approach the true value at the center. Additionally, it is guessed that the values are distributed at regular intervals, and that there will always be a representative of these values which is less and less relevant as we stray from theoretical perfection. Continuity and representativeness of the values condition the choice of a center that will have legitimacy whatever the situation.</p><p>We are reasoning here on a single axis, but if we wanted to cross several categories of data, this central limit theorem would still have to be valid; without it, the machine could not learn, since learning consists of finding the ideal value in a given context. Contextualizing, in turn, is first identifying how far we are from the center, and then choosing the corresponding value, alongside. All situations are representations that relate to the idealized limit, which confers their truth to the contexts of lower and devalued ranks. In a word, learning — with weights — is evaluating.</p><p>If there was no continuity of values relative to the center, we would no longer have the capacity to learn from different contexts because we would be in an exceptional situation. The exception, or irregularity, is what goes beyond the limits that can be evaluated in the context, which theoretically originates from the central value. In truth, we are entitled to ask ourselves: why does centrality determine the contextualization so much? It is that no matter the amplitude of the values, there exists in situ, punctually, only one form of Quality, only one context at the center; all the other samples of contexts represent quantitative variations of the same ideal, or central, Quality. To better understand how ideality and centrality are synonymous, let us move on to the following concept…</p><h4>C. Generalization</h4><p>When building a learning model, we expect it to cover the maximum number of observable phenomena. Let us recall that the model was already established from previous observations; as a result, if the initial frame of reference was respected very closely, we would only be able to learn what scrupulously fits into this reference system, and any new variation would be an unassimilable exception! The learning model will have to budge somewhat from the initial frame of reference; this distancing, which cannot be arbitrary, must also be measured; and it is there that the notions of ideality and centrality come into play: the model is the idealization of a reference system, and the reference system is the center from which the model is built.</p><p>The initial frame of reference lists a set of more or less ideal values, and by transforming it into a model, we recenter all the probable values around the one reference system. In essence, to generalize is to stretch a reference system by broadening its applicability such that it thus imposes itself as a model; which means progressively adding or removing a certain quantity from the initial values of the frame of reference, so that they all become central to it, and that new values encountered can, more or less — or ideally, relate to them. A quality in a model is consequently limited to a standard of minimum and maximum thresholds, in which all values not yet observed could be inserted.</p><p>Let us nevertheless specify that a model most often sticks together variations of many qualities, and that these are most of the time correlated; to this place, the generalization of the initial reference into a model requires adding or removing weight on all the values by considering them all together at the same time — this is when the calculation difficulties arise! Adjusting the weights of the model is therefore only done by iterative graduation, which intrinsically makes the learning possible, a process called “training”.</p><p>The generality of a model is the result of the training phase, and we safely verify that the model is sufficiently general if it learns from data that were not initially located in the frame of reference used for training. Nonetheless, it should not be too general either, otherwise it would lack precision in the information it provides us: a model that does not discriminate anything, that teaches everything about everything, teaches us nothing at all! Neither too much nor too little, these are the evaluation criteria that give us to assess the accuracy of the model; which narrows down to weighting the values of the reference system to the point where they always delimit the unknown values to come; the ideal of the qualities of a model being on balance that they are set at the center of all probable values.</p><h4>D. Latent Space</h4><p>If the qualities that make up a model are reducible to ideal values, the model space, more generally, is only a sheaf of values full of holes to be completed. What is aptly called the latent space weaves series of variations where the missing values will be filled by the integration of data from other situations. Each situation, being at the crossroads of the lines of variations, will complete the latent space to the extent that it fits into one of the unoccupied boxes, at least at this stage of deployment of the model’s latency.</p><p>In summary, learning boils down to indexing all observed values in the latent space framework, which predicts in advance the position of all probable values in conformity with the different dimensions it arranges. A value will never be more exceptional as a function of each of the dimensions, the model integrating it at the right index; and if this does not correspond with the other values of the other dimensions, learning will require amending the data by removing the part that exceeds the limits, and which we easily call: “error”.</p><p>What the error reveals indeed is less the illegitimacy of the model than the irrelevance of the new data that overflows the latent space. Conversely, correcting the error guarantees that the latent space stays afloat despite its occasional questioning; the error should then, for its part, be integrated as one of the variations enclosed in the ideality of the model. The error is basically only what the model lacks in order to certify the stability of the indexes that it has spread over, the variations continuing to be predictable and theoretically calculable, by completeness. In other words, for all situations, the latent space is precisely supposed to be totally graduated…</p><p>This is where this subtle distinction between Graduation and Creation makes sense; everything hinges on the nature of the space in which AI evolves. Relying on data from traces of human experience: images, videos, texts… which recount the world as an object of description, AI only asserts them in a great Graduation where each piece of data regains its latent place. Quite the reverse, an AI that would function by Creation would trace new Perspectives — no longer starting from data, but rather from emerging Ideas. The ideality of the traditional model of Connectionist AI is limited to the Graduation of data captured from the old world; whereas Ideas, laid out by a Creation forever open, will renew the very Quality of experience, offering new Possibilities of Worlds.</p><h3>III. Creative</h3><h4>A. Correction</h4><p>Let us begin again, there are at least two ways for Humans to learn: by doing it themselves and referring to their mistakes, or by imitating in a way that is suggested to them. The Machine also proves capable of learning by availing these two approaches: either we designate a target for it to reach and, if it makes a mistake, it applies a Correction to its errors, until it perfectly achieves what is asked of it; or, another little-used learning mode, we show it by “Porrection” a method to handle, and that the AI will implement as it sees fit.</p><p>Error correction points to a well-known logic problem: XOR — either A or B except both at the same time. The problem is that to encode XOR, you cannot appeal to a single-output function, as you may for AND or OR:</p><ul><li>If A and B are input, AND is true as output;</li><li>If A or B is input, OR is true at output;</li></ul><p>Definitely, if A or B were indifferently at the input, XOR would risk being false for the reason there could be A AND B, which contradicts XOR. This is the reason XOR entails to redouble the function by having both an output which verifies OR on one side, and a distinct one which checks NO-AND on the other side, these two having to be true at the same time.</p><p>Through XOR, AI has a tool to correct its errors, to wit, to check whether two things are resolutely distinct, and if not, it knows in accordance with what threshold it will separate A and B. This XOR technique is the basis of the connection between A and B, thanks to the weight that fixes them on both sides; so that we will find how to go back from the output that gives a function as true, and by “backpropagation” of the error, nuance the inputs A or B upon whether they make the output more or less exact.</p><p>***</p><p>We propose here a learning technique — it is announced: innovative, that we call by Porrection, and which is based on an operator that we invented for this case, the MAND:</p><p><strong><em>Even if A or NOT-A as input, even if B or NOT-B as input, AND is true as output.</em></strong></p><p>We might believe we recognize the famous principle of Antiquity: “ex falso quodlibet” — from the false anything follows; in reality, it is verily from the truth that we start, less as to the conjunction of the truth of A and the truth of B taken separately, than from the truth about the AND of A and B “regarded” together.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*V5JHQ1jyCA1zOlM2tZNpxQ.png" /></figure><p>What we are porrecting thus, is the Regard that holds A and B together, whatever their values; and this is also how we learn most often: we do well and badly simultaneously, besides the fact that the result is sometimes good, despite the errors that are at their origin and their impacts. In learning, we are not going to correct these errors by looking at ourselves; in contrast, we will seek to have the ideal practice be porrected, and thereby regard it better with a view to progressing.</p><h4>B. Composition</h4><p>Learning by correcting its mistakes allows an AI to later on randomly generate an impromptu Composition, which bundles more or less unexpected connections, on top of relative flexibility. Given that the weights applied to each connection are multiple and variable in concert, the generated work will always be different, forming the impression of an unusual composition, since it no longer adequately corresponds to the initial data.</p><p>By porrecting a Regard, it is not the output that is shifted in relation to the input — because of a haphazard treatment; on the contrary, it is the process that is singular, for an output that canonically corresponds to the data entered. It is here that the data are creative — and become Ideas: we no longer aim to recompose them by keeping them separate, but to create a new “Layout”, with the objective that they are held <em>within </em>a single Regard.</p><p>Let us explain how these two approaches of learning by Correction of Composition on the one hand, and Porrection of Layout on the other hand, are sharply opposed; where all the Intelligence comes from the logical combination of AND and XOR, with¹:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/900/1*YG0xhC5aJ9VT6X2PxX1ayQ.png" /></figure><p>¹ <em>A AND B implies C; congruent with: A OR-exclusively B implies NOT-C / A OR-exclusively B implies C; congruent with: A AND B implies NOT-C.</em></p><p>…we note two inconsistencies:</p><p>A ∧ B → C ≢ A ∧ B → ¬ C</p><p>A W B → C ≢ A W B → ¬ C</p><p>We should see here only an exemplification of the principle of noncontradiction:</p><p>¬ ( P ∧ ¬ P )</p><p>…traditionally associated with that of the excluded middle:</p><p>⊤ ( P ∨ ¬ P )</p><p>This dual mode of inclusion associated with exclusion relies on a certain notion of the truth of a statement (⊤) that must be contested; and more precisely, its cornerstone is a conception of reality that isolates too conveniently: “There is…” and “That exists…”. Let us try to join them together:</p><p>⊤ : ◊ ( ◊ ) ∥ ¬ ◊ ( ¬ ◊ )</p><p>⊤ ≢ ¬ ◊</p><p>“There is…”: to assert that it could always be possible, when it is possibly, comes down to saying that it is impossible for it not to be possible.</p><p>In one term, there will be no Impossibility.</p><p>⊤ : ( □ ) □ &amp; ( ¬ □ ) ¬ □</p><p>⊤ ≢ ( ¬ □ )</p><p>“That exists…”: to indicate that it is necessarily, if it must be necessary, amounts to stating that it is necessary that there be no contingency.</p><p>In short, there will never exist a Contingent.</p><p>With the help of the MAND, we conceive a form of Reversibility of both Possibility and Necessity, which authorizes that there is Impossibility just as much as there exists Contingent:</p><p><strong>A ♡ B → C : ◊ (A ∧ B → C) ≡ □ (A W B → C)</strong></p><p><strong><em>That A and B love C is expressed in the fact that it is possible for A and B to give birth to C together, and that it is then necessary for either A or B to raise C as a duo</em></strong>².</p><p>² <em>If you wanted the translation of this sentence in the mediation of a less carnal mathematical formalism, the author would refer you to the next presentation which is to be published, in order to satisfy the more sober taste of the experts…</em></p><p>Conjuncting, exclusively, the Impossibility and the Contingent paves the way for another approach to the problem of Multicollinearity: when data influence an observation very equally and it is necessary to distinguish them to decide which contributes to which effect, we are provoking alternatives of Perspectives by bringing a renewal of Ideas to the situation. These Ideas are leading, by pushing them to the Impossibility, each variable to reveal its true contribution; in parallel, the Contingent of the same Ideas is conducting such and such a contribution interior or exterior to such and such a Quality, instead of varying their effect more or less.</p><p>It has been admitted: by proceeding with the MAND, we maintain the presence of all the data which, in Regard of what emerges, converge towards a new deal; and reciprocally, to every data relate, in emergence, the nascent Ideas — from which all Creativity springs! If an AI that composes tries to bring back to zero the gap to the model to be generated — with the purpose that it is known and acknowledged, Risen Learning — an AI that <em>creates — </em>“counts as one”, to use Alain Badiou’s magnificent words, all differences of the Layout; at the heart of which Layout, it is now only a question of the interiority and exteriority of the Emergence: that is to say of what makes one proper Quality unitarily <em>present</em>, still making many Qualities <em>converge </em>distinctly<em>.</em></p><blockquote>Making emerge the Intelligence of Risen Learning, all things considered, is to succeed in “gluing”, in the sense of Grothendieck, these two schemas put together:</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/333/1*X-kuMr95AYLwKkLluZZXqA.gif" /><figcaption>Presence — Porrecting Ideas towards Qualities</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/335/1*LMCev2FJ6ZHs2bCg6pU4yw.gif" /><figcaption>Convergence — Regarding Ideas interior to Qualities</figcaption></figure><p>The distinction between Quality and Quantity is, after all, the emerging fruit of the “space” where Intelligence happens: is it a question of correcting data, according to their probability, and artificially automating? Differently, are we trying to porrect new possible Ideas, with the intention of augmenting humanly? For the first case, learning is intrinsic to the model, from which automation is launched and relaunched recursively, without enabling any new input from the outside world; in the second, the perspectives drawn give place to a reversible program within which one is always free to manually augment the field of possibilities…</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ed6NgvJ7VL-HI2O8pYsASw.png" /></figure><p>Through each of these two approaches, we surely make use of programming and learning, insomuch as that their subsequent resorption lastly settles which way Intelligence leans: into the automated artifice of Correction / towards an augmentative human Porrection? In conclusion, will the Machine end up learning from all the tasks that humans were engaged in, or will it rather be Humanity that will discover in programming the final meaning of its works? Attempting to answer this question, what is called here Risen Learning is an emerging species of Artificial Intelligence that gives back control to Humans, by reattributing to Ideas the key role of porrecting to Machines.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b8634231469e" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>