Why machines don’t create value (audio of talk)

~1 hour of audio is here

(Invited talk for the Oxford Communist Corresponding Society given on Nov 26th 2020.)

A popular objection to Marx’s labour theory of value is that human labour alone doesn’t create profits: labour must be mixed with capital and land to produce useful output; and, anyway, machines can replicate many tasks that humans perform — there’s nothing special about human labour. Take a human taxi driver. Replace them with a robot taxi driver (some future version of self-driving cars). There’s no difference. The robot passes a Turing test for being a taxi driver. And the company still makes a profit. Hence, human labour is not the sole cause of profit, and the labour theory of value is false.

In this talk I explain why this view is wrong, but nonetheless gives us a useful entry point for gaining a deeper understanding of Marx’s theory of surplus-value, and why the origin of profit is human labour alone.

In the talk I (i) briefly review Marx’s theory of surplus-value, (ii) explain why materialists accept that, in principle, all human capabilities can be automated, (iii) describe a Turing Test for Marx’s theory of surplus-value, (iv) emphasise the importance of understanding that Marx’s theory of surplus-value is irreducibly dynamic in nature, (v) discuss the ideological inversion that causes people to think that machines can create value, (vi) discuss the macroeconomic data that reveals a clear empirical signature that labour creates profit, and (vii) close with some speculations on what may happen when humanity, driven by the blind imperatives of the alien demiurge, is finally forced to abolish itself.

Have a dark but merry Christmas!



  1. Hi Ian! I haven’t gotten around to listening to the full talk, but I have one question. How does the idea that value is slowly transferred from machinery to final commodities apply to digital software? Since software can theoretically last forever, how does it transfer its value to final goods?

    Liked by 1 person

    1. Hi Anon,

      Thanks for your question!

      You asked:

      > How does the idea that value is slowly transferred from machinery to final
      > commodities apply to digital software? Since software can theoretically last
      > forever, how does it transfer its value to final goods?

      The short answer is the labour-value of digital software, when used as constant capital, is transferred to final goods in exactly the same manner as any other element of constant capital. In other words, Marx’s analysis applies without modification.

      The long answer involves explaining why …

      First, why do we care about labour-values at all? Marx intends to capture the real material costs of producing things in terms of human activity. This is of interest and political significance in its own right. Additionally, given capitalist competition, labour-values function as an attractor for market prices, and partially explain their movement over time and relative structure (see this discussion https://ianwrightsite.wordpress.com/2021/03/03/ontocast-interview-on-the-theory-of-the-general-law-of-value/ on the dynamics of the law of value).

      So what is a labour-value? The labour-value of a commodity is (i) a property of the current techniques of production that prevail, not in an individual firm, but in the economy as a whole, and (ii) measures the total coexisting labour that would be required to reproduce a commodity, where reproduction means not only producing the final commodity but also replacing all the means of production used-up both directly and indirectly. A labour-value is therefore a vertically-integrated property of the current techniques of production. (See also https://ianwrightsite.wordpress.com/2017/03/23/coexisting-labour-the-substance-of-marxs-labour-values/ and https://ianwrightsite.wordpress.com/2017/02/28/vertical-integration-and-the-problem-of-deciding-how-much-to-produce/ and https://ianwrightsite.wordpress.com/2017/05/19/just-what-is-a-labour-value-anyway/).

      Marx, in Volume 1, states that workers, when producing a specific commodity, β€œtransfer” the labour-value of any used-up constant capital to the output. Marx’s talk of β€œtransfer” achieves two aims: first, it emphasises that labour-values emerge from the actual activity in production; and second, it illustrates the vertically-integrated nature of a labour-value (e.g. the labour-value of indirect inputs, when they are productively consumed, β€œreappear” in the output).

      Some types of constant capital, such as raw materials, are entirely used-up in the production of a single commodity, whereas other types, such as machinery, persist and participate in the production of huge numbers of commodities. Hence the classical distinction between circulating capital (used-up immediately) and fixed capital (used-up over much longer time periods). As Marx pointed out, the binary distinction between fixed and circulating capital is really shorthand for a spectrum of productive lifetimes of different kinds of constant capital.

      Let’s assume, for simplicity, that the techniques of production, across the entire economy, do not change (assumption of a fixed technique), and that firms in the same sector use identical techniques (assumption of homogeneous technique). If all capital is circulating capital, and the circulating capital is in fact replaced during production, then a labour-value will coincide with actual labouring activities, specifically all the labour performed simultaneously, in different sectors of the economy, that presently are engaged in producing the commodity and replacing used-up means of production. But in general this is not the case. For example, some sectors may draw-down on existing stocks of circulating capital, rather than actively producing replacements. In such circumstances, labour-values will not strictly coincide with actual labour being performed. This separation of labour-values (which are a β€œfield property” of the technical conditions of production; see Ch. 6 of my thesis https://eastsidemarxism.files.wordpress.com/2017/04/wright-thesis-deposited.pdf) from actual labour performed (which depends on how existing capital stocks are utilised in the economy) becomes more pronounced when we consider fixed capital.

      Marx, in Ch. 8 of Vol. 1, gives the example of a machine that spins cotton. Assume that 1000 hours of labour produced this machine. The machine deteriorates, due to wear-and-tear, and finally, after 10 years, irretrievably breaks-down. (We will ignore any labour supplied to maintain and repair the machine during its lifetime). According to Marx, fractions of the 1000 hours of labour-value transfer, over the 10 year period, from the machine to all the commodities it helps to produce.

      In consequence, the presence of fixed capital might seem to imply that the labour-value of a commodity is not just the present labour supplied to produce it, but also includes fractions of past labour that produced the fixed capital, which is now physically embodied in the machine. But a labour-value is neither past or present labour performed, and neither is it literally β€œembodied” in the physical substance of capital. A labour-value is a property of the current techniques of production, and only coincides with current labour performed in special circumstances, and is entirely independent of the level and composition of the capital stock, and whether that capital stock is being actively replaced or drawn down.

      Marx’s talk of the β€œtransfer” of fractions of labour-value from machine to product merely indicates that, given a technique of production that uses this type of machine, then a definite quantity of labour would need to be supplied to replace the fraction of the machine’s use-value that is productively consumed. The labour-value of a commodity includes the replacement cost of these used-up fractions of fixed capital. (Although we may only get to know the average fraction used-up once the machine stops working, nonetheless there is such a fraction used-up before that event. Marx grapples with this epistemological problem via his analogy with β€œlife insurance” in Ch. 8).

      Marx is clear that the labour-value transferred from constant capital is not a function of past labour but the current conditions of production. In the same chapter he gives an example of a change in technique: β€œThe definition of constant capital given above by no means excludes the possibility of a change of [labour-]value in its elements.” And then explains that: β€œIf the time socially necessary for the production of any commodity alters … all previously existing commodities of the same class are affected, because they are, as it were, only individuals of the species, and their value at any given time is measured by the labour socially necessary, i.e., by the labour necessary for their production under the then existing social conditions.” In consequence, if we drop our assumption of fixed and homogeneous techniques, then a situation may arise where a new more efficient machine is widely adopted, yet some firms still use the old, less efficient machine (still operating during its 10 year lifetime). The old machines will continue to transfer fractions of labour-value, but that fraction will now be a property of how the heterogeneous techniques of production distribute across the sector as a whole (i.e., the socially necessary labour-time). And, therefore, an old machine will not transfer the entire 1000 hours of labour-value originally supplied to make it, but transfer less labour-value due to the changes in technique that occurred during its lifetime. (This is not spooky action-at-a-distance but another example of how Marx’s labour-values are properties of a technology field).

      OK, now that we understand how the labour-process β€œtransfers” the labour-value of fixed capital, we can turn to your example of a digital commodity, such as software, that functions as constant capital. Let’s make this concrete, and imagine a commercial 3D modelling tool, such as Autodesk’s Maya, which is used in many different kinds of production processes.

      Software is literally a virtual machine. And virtual machines, unlike physical machines, can be digitally reproduced at (almost) zero cost. This property of virtual machines has led many to assert that Marx’s theory of value no longer applies in the β€œdigital economy”. But typically these assertions are based on a superficial understanding of Marx’s theory.

      The labour-process that produces the first copy of a virtual machine (e.g. many person-years of software engineers constructing source code) is completely different to the labour-process that produces subsequent copies (e.g., largely automated distribution of copies from the cloud). The labour-process that produces the first copy of a virtual machine immediately renders itself obsolete. It happens to be a type of technique that, when performed, necessarily abolishes itself. This is not an entirely new phenomenon, nor an exclusive property of digital commodities, because β€œprinting-like” technologies, where an initial copy takes high effort but subsequent copies are low effort, have been with us since classical antiquity (e.g. see Ch. 3 on Labour Productivity in Classical Econophysics, especially the discussion of early printing techniques and use of moulds: https://bunkerchan.net/.media/692491f5c6a2528c7bbc14581aae2ca84bf72ec00ae833392bcd1d47ccc2725d.pdf).

      A virtual machine therefore has the property that its labour-value is (almost) zero, even when large quantities of labour-time were initially supplied to produce it. (In reality, distributing copies of software to customers incurs low but non-zero real costs. But, for conceptual clarity, we can assume that the labour-value of a virtual machine, as soon as the first copy is made, reduces to zero.)

      This poses a problem for capitalist firms wishing to monetize software and recoup their costs. Hence the proliferation of anti-copying technology that attempts to resolve the contradiction between the forces of production and the prevailing social relations. (Many people now accept as natural that digital goods must be paid for despite their almost zero cost of reproduction. And some of the Utopian ideals of the early internet have faded from memory).

      Marx’s image of labour-value β€œslowly” transferring from machinery to the output is predicated on the use-value of the machine deteriorating through use. The use-value of virtual machines does deteriorate through use but it’s (almost) negligible because it consists in the entropic breakdown of information stored on digital storage devices (e.g. hard drives) due to repeated reads. So, as you point out, we may as well consider the lifetime of a virtual machine to be infinite.

      In consequence, a virtual machine, when it functions as constant capital, transfers 0 labour-value over its infinite lifetime. So digital commodities may last an infinite amount of time, but they are not therefore a source of infinite labour-value. More accurately, a virtual machine transfers fractions of a small quantity of labour-value (the labour-value of creating this particular copy) to all the outputs it helps produce over its (long, but finite) lifetime. So the labour process β€œtransfers” the labour-value of virtual machines in exactly the same manner as physical machines.

      So why do digital commodities have a non-zero price if their labour-value is (almost) zero? Or, more plainly why does Autodesk’s Maya cost over a thousand dollars? Labour-values function as an attractor for market prices, and therefore always affect their trajectory over time, but the attractor state can only empirically manifest if certain conditions are met. One of those conditions, as Marx points out in Vol. 3 Ch. 10, is that different firms are able to compete to supply the same commodity type, and therefore no artificial monopolies exist that can fix prices (see also this paper http://gesd.free.fr/wrightrpe.pdf). Most markets for digital commodities do involve artificial monopolies maintained by property laws. For example, arbitrary firms cannot compete to distribute copies of Maya because Autodesk owns it. Only Autodesk has the property right to distribute copies, and this right is enforced by law and ultimately the power of the state. Counterfactually, if competing firms could enter the market and distribute copies of Maya then the law of value would be free to manifest, and the prices of those copies would gravitate to the (lower) price more closely correlated with the (low) labour-value of producing and distributing a digital copy. But in practice, under our current social relations, this does not happen.

      Your question, as an aside, reveals the importance of not conflating labour-value with exchange-value, as many Marxist interpreters unfortunately do. Of course, money is changing hands all the time. And so constant capital has a monetary representation etc. But when Marx refers to β€œvalue” he does not mean exchange-value but β€œlabour-value”, and he is concerned with the labour process, with the actual material difficulty of making things. And this labour process can very much contradict Capital’s imperative to extract monetary profits from everything we do.

      Thanks for your question, and sorry for the long delay in replying.

      Best wishes,


      1. Thank you for your reply! I had never really though about labor processes making themselves obsolete, so that was certainly a surprising revelation!


      2. Just wanted to ask a quick follow up question, does the higher value of aged wine derive from a transfer of value from barrels to the wine? Thanks

        Liked by 1 person

  2. Hi Anonymous,

    Thanks for your follow-up question.

    The problem of the value of aged wine is a famous thought experiment in classical economics that appears to demonstrate that (any version) of a labour theory of value is false.

    Ricardo, for example, wrote: “I cannot get over the difficulty of the wine which is kept in the cellar for three or four years … which perhaps originally had not 2 shillings expended on it in the way of labour, and yet comes to be worth Β£100”. Since by definition no additional labour is applied during fermentation the increase in value appears to be nothing to do with the labour expended on the product, but rather compensation for the time that the initial investment of 2 shillings of money-capital is β€˜locked up’ in the form of ageing wine.

    Marx mentions, in Theories of Surplus Value, that the problem of aged wine, and related problems, led to the collapse of the Ricardian theory of value.

    The wine thought experiment is simply a different way of expressing the same underlying contradiction that motivates Marx to develop his theory of the transformation in Volume 3.

    Marx initially assumes identical turnover times of individual money-capitals in his presentation of the transformation. He assumes that each individual investment is “tied-up” in production for the same period, say 1 year, until the output can be sold in the market. The only relevant difference between the individual money-capitals is that they are used to purchase different proportions of constant and variable capital (i.e. different processes have different organic compositions of capital). Since profit, according to the labour theory of value, is generated by variable capital alone, then equal investments of money-capital will nonetheless generate different profits because they set in motion different quantities of labour-power. In consequence, the labour theory of value appears to contradict the law of uniform profit-rates, which states that equal investments of money-capital should yield the same profit-rate.

    Marx proposed to resolve this contradiction by distinguishing between surplus-value (that is produced by workers in the process of production) and profit (which is the subsequent distribution of the surplus-value, in the form of monetary profit, to the different money-capitals). Marx’s theory of the transformation attempts to demonstrate that profits are nothing but redistributed surplus-value. And therefore the contradiction between the law of uniform profit-rates and the origin of profit in human labour is merely an apparent contradiction.

    The wine thought experiment makes different assumptions to make the same point. Here we assume that individual money-capitals are invested in production processes with identical organic compositions of capital, but the turnover rates differ.

    For simplicity, assume all production processes require variable capital only. Consider a capitalist A who invests M dollars to employ L workers that produce goods ready for sale after 1 year. After 1 year the L workers have produced S surplus-value.

    In contrast, consider a capitalist B who invests M dollars in wine production that employs L workers for 1 year, but who must then wait 3 years for the wine to age. This process also generates S surplus-value.

    The law of uniform profit-rates implies that capitalists A and B both receive the same profit return, regardless of turnover time. If this were not the case then capitalists would never invest in production processes that turnover infrequently, such as wine production. So capitalist A receives compounded profit over 3 years, and capitalist B, when they sell their wine, also receives a profit equivalent to A’s compounded profit. As Ricardo notes, B’s profit therefore seems to be a reward for simply waiting, and not due to the expenditure of any labour.

    And so we have the same problem again: the law of uniform profit-rates appears to contradict the labour theory of value (the source of profit in labour alone).

    And so back to your question:

    > does the higher value of aged wine derive from a transfer of value from barrels to the wine?

    And the simple answer is: no. The value of the barrels (constant capital) does “transfer” to the wine. But this isn’t the explanation for its “higher value”. Marx’s explanation for the “higher value” of aged wine, despite having no labour directly applied to it, is his Vol 3 theory of the transformation of values to prices of production, which I’ve written extensively about elsewhere.

    Best wishes!


  3. I’d like to confirm my understanding of this. So in the scenario you provided, would the profit rate (ignoring equalization) of the production process of capitalist A be 3x that of the production process of capitalist B? Also, does the higher price of the wine relative to its cost of production come from the fact that the equalized profit rate is higher than the “pre-equalization” profit rate of its process of production?


    1. Hi John,

      Thanks for your question.

      In the example I’m assuming that capitalist A and B receive the same annual profit-rate on their investment of money-capital. Capitalist A gets r% each time from their 1 year investment of M dollars. And capitalist B gets r% compounded over their 3 year investment of M dollars. In other words, given the law of uniform profit-rates, there’s no possible arbitrage between the two production processes.

      And then the problem for the classical labour theory of value arises: the profit-rates are the same but the surplus-value generated by process A is greater over the 3 year period (due to L workers supplying labour for 3 years) compared to process B (due to L workers supplying labour for only 1 year). In order to claim that profit is nothing but surplus-value then this mismatch needs to be explained.

      Hope this clarifies the intent of the example.

      Best wishes,


  4. I’m not sure I’m satisfied with the conclusion of the thought experiment. It’s true that a simple algorithm taxi driver can only be constant capital, as it will always and only function in one way. But with the development of machine learning research that AI can continue to develop itself for as long as desired. While normally an AI is developed to the point where it can consistently perform a designated task, nothing stops you from continuing the learning process into its deployment. I see no reason why eventually there can’t be an AI taxi that finds new means to create value. As such, what once was simply constant capital is now artificial variable capital.


    1. Hi Xeno,

      Thanks for your question!

      > I see no reason why eventually there can’t be an AI taxi that finds new means to create value.

      I completely agree!

      > As such, what once was simply constant capital is now artificial variable capital.

      No. All that’s happened is that yet another activity that once could only be performed by variable capital can now be performed by constant capital (due to technological progress). A more sophisticated robot taxi driver, now able to notice new market opportunities in the domain of road transport, still lacks the universal causal powers of variable capital. The more sophisticated robot cannot work harder or smarter, in every production process, to change the conditions of production and cause changes in the level of profits (as per the “causal powers” thesis).

      Another reader had a similar response to yours, and I replied in more details in cosmonaut mag:

      Hope this helps!

      Best wishes,


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s