<M>OR<H>

A

We can have transparent and trusted digital information by clearly tagging the origin of content as Machine or Human, enhancing the benefits of AI for Humanity by reducing the risks.

B

This is a not-for-profit solution proposed by Matt Kuperholz, an Australian AI practitioner who has been working in the field for over 25 years and has been focused on responsible and ethical aspects of AI development and regulation since 2018.

C

You can understand <M>OR<H> in as much detail as you need through progressive story-telling by scrolling as far down as you want through this webpage, or you view the entire content all at once. There is also this video overview and you can skip straight to this call to action at the end.

A

Competitive and commercial forces relentlessly drive progress without pausing to solve for the risks. The exponential speed of development makes many specific guardrails quickly irrelevant, unsuitable and ineffective.

B

It is becoming increasingly difficult to tell Machine generated content from Human generated content. Soon it will be impossible. We will be unable to tell a real photograph, phone call, photograph, social media post or article from a fake one. This will harm both Machines and Humans!

C

<M>OR<H> immediately provides Humanity a simple and flexible solution for foundational problems of trust introduced by AI, allowing Machines and Humans to be certain of answers to important questions including : Is this content generated by Machine or by direct Human agency? How do we verify by what Machine or by which Human? Is this the original content or has it been modified? How can we trace, replicate and audit the Machine content?

A

Humans will be able to verify what is Human generated content (and prove which Human published it if they choose to publish that).

B

Humans will be able to verify what is Machine generated information (and prove which Machine generated it).

C

We already use watermarks to verify the authenticity of currency and information contained within official documents. They are widely recognised as a way of proving legitimacy. We already use nutrition labels to provide transparency about what is in our food. Similarly, the <M>OR<H> solution can provide transparency and increase trust about the origins of our information diet for Humans.

A

The simplest approach is to agree we will tag content as <M> or <H> and that we decide to only trust information that is tagged accordingly. Non-tagged information should be considered less trustworthy! There are already many existing great tools, standards, protocols and ideas for addressing different aspects of this challenge - <M>OR<H> provides a straightforward mechanism for unifying these different approaches.

B

These tags are infinitely extendable with their attributes : included inside the tags so we know it actually is a Machine or Human publisher, using the same Digital Certificates that places the padlock icon in our browser so that we can trust we are really interacting with our bank, our email provider or the genuine social media website. An ecosystem of Trusted Certificate Authorities already exists that can be simply integrated with <M>OR<H> to prevent fraud that will allow us to trust the origin of the content as Machine (and which version of which model with what parameters) or Human.

C

Other information can be included in the attributes of the <M> or <H> tags so that we can trust the information has not been tampered with or altered, using the same Digital Signatures that verify the authenticity of legal contracts or financial statements. For example, a document signed with a digital signature will verify the signer’s identity and that the content has not been altered since it was signed. This proves that the content is authentic, providing a reliable method for verifying the origin and integrity of both Human and Machine-generated information.

A

Popular phone, browser and computer operating systems and vendors like Google, Apple and Microsoft already address required security, identification and authentication which could be easily extended to automatically place more attributes into the metadata of content created using those ecosystems.

B

For commercial and reputation protection and enhancement, reputable and trusted AI service providers can share the model version, creator and other key variables in attributes of the <M> tag to facilitate traceability, replicability, auditability; providing trust and governance by publishing how their AI has been assessed for bias, robustness, security in accordance with accepted standards.

C

It is relatively straightforward to agree to do <M>OR<H> and to start doing it. This immediacy is deliberate by design as other practical aspects for regulating, producing or using AI to ensure it is trusted and responsible are relatively complicated and unclear and we (Humans) need to accelerate our efforts. Responsible AI performance metrics and jurisdictional regulatory requirements can be continually integrated into <M>OR<H> as they develop.